Home

Classification, Regression, Segmentation and Clustering

1. Explanatory variables see page 278 Weight variables see page 293 root Terminological morpheme that is used alone as word root word or as basic element in a derived word Rule mode A simplified Classification Regression mode that is used to express the model with rules S Score The numeric evaluation mark in view of a given problem scorecard This screen provides you with the coefficients associated to each category for all variables in the model only in case of aregression model Classification Regression CUSTOMER SAP Infinitelnsight 7 0 287 2014 SAP AG or an SAP affiliate company All rights reserved Glossary seasonal Variations due to calendar events sensitivity Sensitivity which appears on the Y axis is the proportion of correctly identified signals true positives found out of all true positives on the validation data set Sequential cutting strategy The sequential strategy cuts the initial data set into three blocks corresponding to the usual cutting proportions The lines corresponding to the first 3 5 of the initial data set are distributed as a block to the estimation data set The lines corresponding to the next 1 5 of the initial data set are distributed as a block to the validation data set The lines corresponding to the final 1 5 of the initial data set are distributed as a block to the test data set session A session is identified by a unique key and is
2. An API also allows you to interface SAP Infinitelnsight with any other application SPSS Microsoft Excel for example and gain access to any other data source You must develop a specific dll script for each new source l Note For information about data formatting and specifically for the list of supported ODBC compatible sources see the document Data Modeling Specification 4 3 Data Set To use SAP Infinitelnsight features you must have a training data set available that contains the target variable with all its values defined Then you can apply the model generated using the training data set to one or more application data sets 4 3 1Training Data Set A training data set is a data set used for generating a model In this set the values of the target variable on page 31 or variable corresponding to your business issue are known By analyzing the training data set SAP Infinitelnsight features will generate a model that allows explanation of the target variable based on the explanatory variables To allow validation of the model generated the training data set is cut into three sub sets using a cutting strategy on page 19 The training data set may correspond to either a complete population section of your database or a sample extracted from this population The choice depends on the type of study to be performed the tools used and the budget allocated to the study CUSTOMER SAP Infinitelnsight 7 0 18
3. E Ds B Stop View Type Copy Print Save Applying the Model 3 Stop Current Task 44 Cancel qf Previous IP Next CUSTOMER SAP Infinitelnsight 7 0 175 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Analyzing the Results of the Application For this Scenario Open the results file in Microsoft Excel in the text format that you obtained when you applied the model to the CensusOl csv file VI To Open the Model Application Results File 1 SS i aM OO al Goad Dal Daali 0 0 00 OO A an A B C KxIndex class rr_class 0 02626554 0266730263 0 192986576 0 027590154 0230136439 035434 1A 0 350695954 0 1586327 27 0 654126425 0 540619075 0 2355657605 0317775369 0 1920477 45 0 11635045 0164729876 0 17731364 0567965505 00237732A 0094321767 0 540656046 Of37949155 0 00095057 0435952127 0 964523262 0 91677 2066 055152601 4 0676 07566 0024259994 0 056997665 0406741321 proba_rr_class bar_rr_class 117675017 49620065 7 0457313470 0 946476877 1521923104 401565995 O 467965275 4052460025 0575066474 0 9330607 06 15194953461 1426705707 04632550654 0731234645 1466572215 000014916 0 00604592 0 00036669 0 00765412 0 00574755 0 00092295 0 00550965 0 007116805 0 00415955 0 00175730 0 OO0S2255 O 0046 7542 0 00642789 000360329 0 00066525 contrib age contrib_wo
4. 159 Note If the Kx ndex variable of the model is virtual the application data set must not contain a physical Kxlndex variable CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification For this Scenario Due to technical constraints a data set corresponding to the database of 1 000 000 customers that will be used in this scenario can not be provided to you You will apply the model to the file CensusOIl csv which you used to generate the model In this manner you will be able to compare the predictions provided by the model to the real values of the target variable Class for each of the observations In the procedure To Apply the Model to a New Data Set Select the format Text files nthe Generate field select the option Individual Contributions Select the folder of your choice in which to save the results file Wodel Generated Output Do not select the option Keep only outliers vV To Apply the Model to a New Data Set 1 On the screen Using the Model click the option Applying the model to a new data set The screen Applying the Model appears a SAP InfiniteInsight VX y 2 class_Census01 File Help Applying the Model Application Data Set Data Type Text Files W Folder Samples Census Data a ia Define Mapping Generation Options Generate Predicted Value Only
5. 4 Save in Variable Pool A ip Analyze H Save Description Remove from Variable Pool View Data Properties Description 1a VIEW Description Desc_Census01 csv Index Name Storage Value Key Order Missing Group Description Structure djage number continuous 0 a 2workclass string nominal 0 0 a 3 fniwat number continuous 0 0 a 4 education string nominal 0 0 a S education n number ordinal 0 0 a 6 marital status string naminal 0 0 a T occupation string nominal 0 0 Fr relationship string nominal 0 0 S grace string nominal 0 0 Z 10 sex string nominal 0 D a 11 capital gain number continuous 0 0 99999 a 12 capitalloss number continuous 0 a 13 hours per w number continuous 0 0 a 14 native country string nominal 0 0 7 15 class number nominal 0 0 a 16 Kxlndex integer continuous 14 0 Automaticall a _ Add Filter in Data Set cs Oy 6 Click Next CUSTOMER SAP Infinitelnsight 7 0 71 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Why Describe the Data Selected In order for the SAP Infinitelnsight features to interpret and analyze your data the data must be described To put it another way the description file must specify the nature of each variable determining their Storage format number number character string string date and time datetime or date date l Note When a variable
6. Predicted Target Category Predicted Non target Category Positive Observations Predicted Negative Observations Predicted True Target Category Number of correctly predicted Number of actual positive observations positive observations that have been predicted negative Actual Positive Observations True Non target Number of actual negative Number of correctly predicted negative observations that have been observations Category predicted positive Actual Negative Observations By default the Tota Population is the number of records in the Validation data set You can modify this number to see the confusion matrix for the population on which you want to apply your model The Metrics The Classification Rate is the percentage of data accurately classified by the model when applied on the training data set The Sensitivity is the percentage of actual positives which are correctly identified as such The Specificity is the percentage of negatives which are correctly identified as such The Precision is the percentage to which repeated measurements under unchanged conditions show the same results The Score indicates how sensitively a likelihood function depends on its parameter The likelihood of a set of parameter values given some observed outcomes is equal to the probability of those observed outcomes given those parameter values CUSTOMER SAP Infinitelnsight 7 0 145 2014 SAP AG or an SAP affiliate company
7. Tech supp Not in fan ir 25 202480 Assoc acdm 12 Never marr Other rela 18 25 Private 332702 Assoc voc 11 Never marr Other seri Own child 19 25 Private 316688 H3S grad 9 Never marr Machine o Not in fan 20 25 Private 499233 H5 grad 9 Divorced Adm clerical Notin fan 21 25 Private 98155 Some colle 10 Never marr Transpor Unmarrie Close VI To Save a Filter You can save the filter you created to be able to reuse it at a later moment without being obliged to recreate the same conditions 1 Click the button Save Filter A pop up window is displayed 2 Inthe list Data Type select the format in which you want to save the filter 3 Use the Browse button located on the right of the Folder field to select the folder or database where you want to save the filter 4 Inthe Description field enter the name of the file or table in which you want to save the filter 5 Click OK vV To Load an Existing Filter To apply a filter to the data set you can use a file created during a previous use of the data set in Infinitelnsight 1 Click the button Load Existing Filter A pop up window is displayed 2 Use the list Data Type to select the format of the filter 3 Use the Browse button located on the right of the Folder field to select the folder or the database in which the filter is stored SAP 4 Use the Browse button located on the right of the Description field to select the file or the table containing
8. You have had significant experience in data modeling You have previously taken a SAP Infinitel nsight training seminar You are already a SAP Infinitel nsight user CUSTOMER Best Use of this Guide You could restrict yourself to 1 Reading the scenario of the feature that interests you or at least the summary of that scenario Application Scenario Enhance Efficiency and Master your Budget using Modeling Application Scenario Customize your Communications using Data Modeling 2 Going directly to the relevant section Infinitel nsight Modeler Regression Classification nfinitel nsight Modeler Segmentation Clustering Read all sections of the guide through at least once in the order in which they are presented In both cases ensure that you have a complete grasp of the essential concepts relating to the use of SAP Infinitel nsight by reading chapter Essential Concepts on page 16 These concepts are essential both for the use of SAP Infinitel nsight features and for analysis of the results obtained You could limit yourself to 1 Verifying that you are familiar with the terminology used by Infinitel nsight by examining the contents of chapter Essential Concepts on page 16 in the detailed table of contents 2 Reading the summary of the scenario of the feature that interests you Application Scenario Enhance Efficiency and Master your Budget using Modeling Application S
9. 75 of the prospects declined your invitation Your business issue consists of understanding the test results by identifying the reasons which led certain individuals to respond favorably to your offer and others to respond in the negative Then you would be able to use the analytical model obtained to predict the behavior of each of the 1 000 000 prospects in your database This would ensure that you optimized your marketing campaign by making the offer only to those individuals most likely to be interested The file containing the data set used for the test was sent to you by the Information Technology department of the bank in the form of a flat file csv This file corresponds to the sample file CensusOl csv provided with SAP Infinitelnsight and described in the Introduction to Sample Files see page 60 section CUSTOMER SAP Infinitelnsight 7 0 56 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 5 1 5 Your Business Issue Following the test phase of your campaign your marketing database will contain A list of 1 000 000 prospects A list of 50 000 prospects selected in a random manner during the test phase and whose response to your test campaign is known This sample taken from your initial database also contains missing values and correlated variables Your approach to the business issue consists of using the data set in its present state as a train
10. CUSTOMER End User Documentation Document Version 1 0 2014 07 1 1 1 2 1 3 2 2 3l ea eS 4 1 4 2 4 3 4 4 4 5 4 6 4 7 FIOW TO Use this Document asiaa S a edueuleeadeate 4 Organization of this DOCUMEN Teansannnnnni e E E E E E E N E A E A N 4 WME eco ONO SAG OEE S 5 Conventions Used in this DOCUMent sssssssseunnnrrrrrrrrssssssesttttttrrrrrtnrr nated eugelud ante cseaee caueede eta ieandenateeceades 6 WeGICOME to this GU eas are iiica tai cinntuescuntanunsnanadendnunnnesvunnaices uianuand usados EA A aA DAAE OE A A AA N aa AANA EAOa 7 PAD OUIE TMS UOC MN TM EM etc hat a A A AA AA A RR R 7 2 1 1 Whe Snould Readthis DOCUMEN orsinil a a i a ROS 7 Zee Prerequisites for Use of this Document 0 ccccccccceeec eee e eee ee eee ee eee eeeeeeeeeeeeeeeeeeeseeseeeseseseeeeeeeseseeeeees 7 Ao Whatta Document COVEN 03 55 51 nN 8 BelOre BeCINNING EE EE E E E E E E E E EE EET 8 Ze Files and Documentation Provided with this Guide 0 0 ccccccsscccceceeeeececceeseeeeseesseaeeeeeeeeeeeeeeeeeettnnaas 8 SAP TTI MSI OIE sec sacnivaneuseevnvenauenseeusdaciotaveus urvwucdasuaausuuusuveautiesdvewsyeusuadevusducnaueudieulvanecsssveedvauavvsduuusstnadensuaiis 10 NITE OCC CIOs eiee a e E E E E E A N EEEN EAE 10 Architecture and ODeralON eisean a a ae bee 11 3 2 1 OCF OE E EE a oe bee ee ose ce bee see eanme et ace 11 3 2 2 OS a OI eana easter ete acer ant atin easiness et ete hee ee iae 12 Methodological Pr
11. Regression Classification Defining the Degree of the Model optional The model generated by Infinitelnsight Modeler Regression Classification is represented by a polynomial This polynomial may be of degree 1 2 3 or greater By defining the polynomial degree you will define the degree of complexity of the model It is greatly recommended that you always use a degree of 1 default value for the first analysis of a data set Using a higher degree of polynomial does not guarantee that you will in all cases obtain a more powerful model For more information about the polynomial degree Representation of a Model see page 36 For this Scenario Keep the polynomial degree set to the default value that is 1 vl To Define the Degree of the Model In the Polynomial degree field enter the value corresponding to the degree of complexity of the model that you want to obtain Polynomial Degree 1 setting the Score Bins Count This option allows you to define the numbers of bins to create for the score This value must be set between 20 and 100 since a lower or higher number of bins would lead to poor model quality Score Bins Count 20 Exclusion of Low KR Variables This option allows you to enable the exclusion of variables based on the value of their prediction confidence KR Infinitelnsight uses an internally computed threshold to decide whether a variable has a low prediction confidence This threshold depends
12. eh aor eer 40 ar Poia on opt op we percentage E Random Wizard Validation The default parameters display the profit curves corresponding to the Validation sub set blue line the hypothetical perfect model Wizard green line and a random model Random red line The default setting for the type of profit parameter is Detected profit and the values of the abscissa are provided in the form of a percentage of the entire data set CUSTOMER SAP Infinitelnsight 7 0 117 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification When the target is continuous the following curve is displayed Debriefing Type Predicted a Vv Models rr_age Vv Actual a 22 5 25 0 27 5 30 0 32 5 35 0 37 5 40 0 42 5 46 0 47 5 50 0 52 5 55 0 57 5 60 0 62 5 Predicted Wizard Validation The default parameters display the curves corresponding to the Validation sub set blue line and the hypothetical perfect model Wizard green line The blue area represents the standard deviation of the current model For more information on the meaning of model curves see Understanding Model Graphs on page 122 When there is more than one target select the target for which you want to see the curves in the Models list l Note To each variable corresponds a model The name of each model is built from the rr_ Robust Regression prefix and the model target name Select
13. the structure The edit window opens If the structure had been extracted from the variable statistics or from a model the fields are already filled amp relationship xi Group Structure Category Edition di Husband Wife i Husband i Wife di Not in family Not in family J Other relative Other relative d Own child Own child J Unmarried Add New Group Add Category Unmarried New Category Remove Group Remove Category Merge Add Missing Alphabetic Sort gt Advanced oK Cancel For details on how to use the Structure editor see Structure by Type of Variables CUSTOMER SAP Infinitelnsight 7 0 81 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Structure for a Continuous Variable The structure for a continuous variable is defined by several intervals each made of a lower bound J that can be either open or closed a minimum value Minimum a maximum value Maximum ahigher bound J that can be either open or closed All intervals must be adjoining there can be no gap or overlap between two intervals The option Add Missing allows you to indicate with which interval the missing values should be grouped The option nclude Smaller Data allows you to include to the first interval any value smaller than its lower bound In the same way the option nclude Higher Data allows you to include to the last
14. 1s The query is accepted and executed immediately Batched ls lt duration lt 2 s The query is accepted but will be executed on next idle time Rejected Zs lt duration The query will never be executed The number names and limits of classes are defined by the user in order for these values to match the current DBMS configuration and DBMS usage policy The Explain Mode has been Configured If the Explain mode has been configured by your DBMS administrator there are two possible outcomes to a query the query is accepted and executed this is completely transparent SAP Infinitelnsight accesses the data without further input from the user the query needs to be validated before being executed a pop up window opens displaying a message configured by the DBMS administrator A query that needs validation can be categorized in two ways medium sized You will probably have to check with your administrator to know which action to take fthe administrator authorizes the query click the Continue button The pop up window closes and the requested action Is carried out f the administrator does not authorize the query click the button Stop Query the pop up window closes but no action is executed huge It means that the query will take too much time and resources In that case the behavior of the Continue button depends on the configuration set by the DBMS Administrator for example it can automatically ref
15. 2 Select the radio button Uniform target encoding Enable post processing Original target encoding Uniform target encoding CUSTOMER SAP Infinitelnsight 7 0 103 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Defining the Target Key Values For the binary targets you have the option to select which value is the key category for each target By default the category selected by SAP Infinitelnsight is the least represented in the data set The Advanced Model Parameters screen lists all the binary targets of the current model allowing you to define the key category for each target that is the expected value of the target In this scenario Do not define a value for the target variable SAP Infinitelnsight will automatically select 1 as the key category for the Class variable T To Define the Key Category Value for a Target Variable In the Target Key field corresponding to the chosen target enter the key value Target Key Settings Target Target Key Ce Auto selection Tab The Auto selection tab allows you to define the parameters of the automatic variable selection a SAP InfiniteInsight Vx y Z class_Census01 File Help Advanced Model Parameters General BAe las Risk Mode Gain Chart will Select the best model keeping between 1 and all variables iEnable Auto selection i Variables Selection Parameter
16. 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 3 2 Application Data Set An application data set is a data set to which you apply a model This data set contains an unknown target variable for which you want to know the value The model applied to the application data set must have been previously generated from a training data set The application data set must contain exactly the same information structure as the corresponding training data set that is The same number of variables The same types of variables The same order of presentation of these variables A Caution The application data set must contain a target variable that corresponds to that of the training data set This point is true for all instances even if the values of this target variable are empty When these values are defined they may serve to detect any possible deviant observations outliers 4 4 Cutting Strategies 4 4 1Definition A cutting strategy is a technique that allows decomposition of a training data set into three distinct sub sets An estimation sub set A validation sub set A test sub set This cutting allows for cross validation of the models generated There are nine types of cutting strategies available within SAP Infinitelnsight CUSTOMER SAP Infinitelnsight 7 0 19 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 4 2 Roles of the Three Sub
17. Box AU Models Are Wrong But Some Are Useful l Note Quote from Robustness is the Strategy of Scientific Model Building in Robustness in Statistics eds R L Launer and G N Wilkinson 1979 Academic Press 4 7 2Performance of a Model A model that has satisfactory performance is one that possesses both High explanatory power that is sufficient capacity to explain the target variable This explanatory power is indicated by the predictive power of the model High robustness that is sufficient capacity to repeat the same performance on new data sets containing observations of a similar nature to the training data set This explanatory power is indicated by the prediction confidence 4 7 3 Types of Models In Data Mining there are two types of models Predictive and explanatory models which allow one to predict and explain phenomena Descriptive models which allow one to describe data sets With SAP Infinitelnsight you can generate models that are both highly descriptive and highly predictive CUSTOMER SAP Infinitelnsight 7 0 35 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 7 4 Generating a Model SAP Infinitelnsight models are generated during a phase called the training phase using a training data set Depending on the situation the training data set may be cut into three sub sets An Estimation sub set A Validation sub set A Test sub set A cutt
18. CUSTOMER SAP Infinitelnsight 7 0 91 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 10 11 Click OK To Save the Categories Translation Translate the variable categories as described above Click the Save button Choose a Data Type Select a Folder Enter a Name for the file or table Click OK To Load an Existing Translation File Click a nominal variable to translate its categories Go to the Edition tab of the ribbon and click the option Translate Categories Anew window appears Click the Load button Select the format of the translation in the list Data Type Use the Browse button located on the right of the Fo der field to select the folder or the database in which the description is stored Use the Browse button located on the right of the field Table or File to select the file or the table containing the description Click OK Click the button no Update to refresh the display of the categories If the list of columns is not named correctly use the Advanced Settings _ a header line and update again see next paragraph to set Map the language names with those from the loaded translation by clicking the categories and choosing the corresponding language in the contextual menu Click OK selecting Variables Once the training data set and its description have been entered you must select the following variables 92 one o
19. Centroid Schema Key Observation Unassigned Observation Overlapping Qhserration __ SaL Defined Cluster Overlap Area centroid Division How to decide which segmentation is better As aside effect of the supervision Infinitelnsight Modeler Segmentation Clustering provides you with a KI and KR It can be used to compare the two segmentations especially because the number of segments is the same If KI does not change significantly then the one with SQL may be preferred because it is easier to understand If there is a fall of KI you may want to stick with the basic segmentation KI may not be the thing you want to optimize for segmentation The target profile of each segment is available in the GUI Out of the four clusters maybe one or two are of real interest In that case you have to focus on these interesting segments and see how they evolve with SQL generation statistical Reports The Statistical Reports provide you with a set of tables that allows you a more detailed debriefing of your model These reports are grouped in different levels of debriefing the Descriptive Statistics which provides the statistics on the variables their categories and the data sets as well as the variables cross statistics with the target l Note If your data set contains date or datetime variables automatically generated variables will appear in the statistical reports For more information refer to section Date and Datetime V
20. I ndicates For instance Cluster The name of the cluster Cluster i Name Frequencies The number of observations contained in The customers contained in cluster 1 represent 4 22 of the the cluster relative to the total number of total number of customers contained in your entire training observations contained in the data set data set of The proportion of observations 51 17 of the customers contained in cluster 1 belong to the contained in the cluster belonging to the target category of the target variable Class In other words target category of the target variable 51 17 of the customers contained in this cluster responded in a positive manner to the test phase of your marketing campaign This table allows you to select the cluster for which you want to view cross Statistics CUSTOMER SAP Infinitelnsight 7 0 254 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Inthe lower part a plot presents either the cross statistics corresponding to the cluster and the variables selected or when it has been calculated the SQL expression defining the cluster The figure below presents the screen Cluster Profiles which appears as the default plot for this scenario The plot presents the SQL expression for cluster 1 295 a SAP InfiniteInsight VX y 2 cass_Census01 Fil aly Cluster Profiles Cluster Name r 2 3 4 5 6 T E a x asad
21. Kxen Simple 1 2004 12 16 K25 Census Class Kxen Simple 1 2004 12 16 K2R_census_3 3 0 Kxen Simple 12005 09 05 veel K25 Census 3 3 0 Kxen Simple 1 2005 09 05 K2R_Census_331 IkKxen Simple e 1 2005 09 08 dass_K2R_Apply Kxen Simple 1 2005 09 08 K2R gt Census gt IKxen Simple 1 2005 09 13 K2S_Unsupervize Ikxen Simple 1 2005 09 20 K2R Census Age Kxen Simple 1 2005 09 21 K2R gt sex gt Cen Kxen Simple 1 2005 09 23 3 3 1 gt census gt Kxen Simple 1 2005 10 17 CUSTOMER 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification E OK Cancel SAP Infinitelnsight 7 0 4 Inthe list Data Type select the type of store the model is saved in 5 Use the Browse button located next to the Folder field to select the folder or database containing the model 6 Inthe displayed models list select the model from which you want to extract the variable structure 7 Click OK 8 Inthe list Target from Loaded Model select the target of the model The variables you have selected are displayed in a list with the corresponding variables from the loaded model X K R_Census_ 331_1 xj Target from Loaded Model Variables from Training Census0 1 csv age Variables from Loaded Model K2R_Census_331_1 age ka oe Add View Census0 Licey K2R_Census_331 Version 1 marital s
22. To use this mode you need to choose a Risk Score associated with a Good Bad Odds ratio l Note The odds ratio is the ratio between good and bad i e 1 p p where p is the probability of risk the number of Points to double the odds l Note The points to double the odd PDO are the number of risk points needed to double the odds ratio CUSTOMER SAP Infinitelnsight 7 0 107 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Example Considering a Risk Score equal to 615 an odds ratio of 9 1 and 15 points to double the odds In this case Infinitelnsight will automatically re scale the internal scores to scores in Risk Mode space and associates an odds ratio to each score in Risk Mode space In this Scenario Do not activate the Risk mode VI To Define the Risk Mode Parameters 1 Inthe field Risk Score enter the score you want to associate with a good bad odds ratio 2 Inthe field for good bad odds ratio of enter the ratio 3 Indicate the increase of score points needed to double the odds in the field Points to double odds 4 Click the button View Score Table to display the table of scores associated with the corresponding good bad odds ratio Risk Score Odds Ratio 585 2 25 600 45 615 9 0 630 18 0 645 36 0 660 T20 675 144 0 690 288 0 705 576 0 720 1152 0 RISK Fitting Domain This option allows the user to control the
23. Type of Results score or predicted value probability prediction range or maximum error individual contributions decision Description For a continuous variable the predicted value corresponds to the value predicted by the model for the target variable of each observation The predicted values correspond to the values read off the X axis of the profit curve plot The predicted value of each observation is calculated by replacing the parameters of the polynomial representing the model by the values of each of the variables of that observation For a binary variable the model outputs a score Corresponds to the probability of each observation belonging or not to the target category of the target variable The prediction range allows you to identify outlier observations An observation is considered an outlier if the difference between its predicted value and its real value exceeds the value of the prediction range In other words the prediction range is a deviation measure of the values aroud the predicted score The individual contributions by variables contained in the data set with respect to the target variable The sum of all those individual contributions corresponds with the predicted value score to the nearest whole number The decision option can only be used for classification models that is when the target variable is nominal It allows to generate a classification decision based on the scores or pr
24. amp View Type Fraction Help Bar Sort Reset Cluster Copy Print Save Export to Orientation Categories Names Excel r Frequencies of T D 27 16 422 BB 5 88 E 9 23 O E ira 4 40 EEE Variables occupation ey v Fix Variable Cluster 1 vs Whole Population for Variable occupation 0 75 0 50 0 25 o 00 i th A ed xl a ath fe e P le a ao ee a a Xy so ia al PEO p en pa Sale oad oF ete et ae Categories E Al Population Cluster 1 CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Cross Statistics Plots Cross statistics plots contain two curves The blue area corresponds to the profile of the variable selected over the cluster selected The red area corresponds to the profile of the variable selected over the entire data set The figure below presents the Cross Statistics obtained in this scenario for cluster 6 a SAP InfiniteInsight VX y 2 cass_Census01 File Help ali Cluster Profiles a y 2 ASW B View Type i Bar sort Reset Cluster Copy Print Save Export to Orientation Categories Names Excel bil Cluster Name Frequencies of T a my t 44 Me 2 25 5 E 11 32 6 5 5 T E 13 75 1 42 8 B 3 61 0 28 9 E 10 32 PE 13 8 10 f 3 93 0 63 EEE Variables penie lS E w A Fix Var
25. and actively orient the training process To declare a variable a weight variable results in creating a number of copies of each of the data set observations proportional to the value they possess for that variable Specifying a weight variable can be used either to assign a higher weight to a single line or to do stratified sampling The effect of the weight can be considered as the following a line with a weight of two in the training data set is exactly equivalent to having two identical lines with a weight of one Example Imagine a data set in which the observations correspond to Individual Americans These observations are described by the variable age among others Defining the variable age as a weight variable means that for generation of the model older individuals will be weighted more heavily than younger individuals Constraints Governing Use Only positive continuous variables may be used as weight variables CUSTOMER SAP Infinitelnsight 7 0 34 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 7 Models 4 7 1Fundamental Definition The term model carries many different meanings depending on its field of application In Data Mining a model describes and explains the relationships which exist between input data explanatory variables and output data one or more target variables It allows one to predict and explain phenomena or to describe them To quote George E P
26. click the Close button 5 2 4 Step 4 Using the Model Once generated a classification model may be saved for later use A classification model may be applied to additional data sets The model thus allows you to perform predictions on these application data sets by predicting the values of a target variable The model can also be used to carry out simulations on specific observations on a case by case basis Moreover you can refine a classification model by re generating it with an optimized list of explanatory variables The SAP Infinitelnsight allows you to select the variables most pertinent to your business issue automatically with pertinence defined as producing the minimum area between the predictive curve and the hypothetical perfect curve and thus maximizing the volume of information explained by the model So that you to apply the model to any other database SAP Infinitelnsight allows you to generate different source code of the model C XML AWK HTML SQL PMML2 SAS JAVA Analyzing Deviations The option Analyze Deviations is a tool that provides you with a diagnostic of the data statistical variation This option can be used for several purposes to compare the distribution of a new data set with the distribution of the data set used to train the model to check the quality of new data after loading them to check if your data have evolved over time and thus if the model need to be adapted to the new
27. i V jis Category Frequencies on Control Group Report Style Automatic use report preferences v 4 Warning Including charts makes saving the reports longer CUSTOMER SAP Infinitelnsight 7 0 158 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification In the displayed list check the sections you want to Save In the list Report Style select the type of output you want Three styles of output are available Automatic saves the default view displayed in the interface Graphical saves the reports as graphs if such a view exists Textual saves the reports as tables Click OK Select the folder in which you want to save the report Enter the name of the file A Caution When selecting the options Automatic or Graphical be careful to choose an appropriate file type such as pdf rtf or HTML Applying the Model to a New Data Set The currently open model may be applied to additional data sets The model allows you to perform predictions using the application data sets and specifically to predict the values of the target variable Constraints of Model Use In order to apply a model to a data set the format of the application data set must be identical to that of the training data set used to generate the model The same target variable in particular must be included in both data sets even if the values are not contained in the application data set
28. pak a i pak at T o gh gh ot aah ag apr age apr ar apr ao wr aan apr aor aan aut att aah oft we percentage E Fandom E Wizard Validation The default parameters display the profit curves corresponding to the Validation sub set blue line the hypothetical perfect model Wizard green line and a random model Random red line The default setting for the type of profit parameter is Detected profit and the values of the abscissa are provided in the form of a percentage of the entire data set 2 When there is more than one target select the target for which you want to see the curves in the Models list l Note To each variable corresponds a model The name of each model is built fromthe kc_ prefix and the model target name 3 Select the viewing options that interest you For more information about viewing options Viewing Options Plot Options vV To Display the Graphs for the Estimation Validation and Test Sub sets Click Data Set and select one of the following options that allow you to switch between the graph for the Validation sub set ert the graphs for all the sub sets 23 l CUSTOMER SAP Infinitelnsight 7 0 236 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering M 237 To Change the View Type Click View Type and select the desired option To Copy the Model Graph Click the Copy button and select the desired option The app
29. the filter 5 Click OK CUSTOMER SAP Infinitelnsight 7 0 216 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Translating the Variable Categories You can translate the categories of a nominal variable save the translation or load an existing translation This translation has no influence on the variable structure which has to be set according to the original values of the variable l Note The variable Target Key that is used in the advanced settings for example does not take into account the translation when displaying the possible values of this variable vV To Translate the Variable Categories 1 Click a nominal variable to translate its categories 2 Goto the Edition tab of the ribbon and click the option Translate Categories Anew window appears 3 Choose into which languages you want to translate By default the language of the user interface is displayed as acolumn 4 Clickthebutton to extract the variable categories from the data set 5 Translate the categories l Note You do not need to fill all fields Click OK To Save the Categories Translation Translate the variable categories as described above Click the Save button Choose a Data Type Select a Folder Enter a Name for the file or table Click OK To Load an Existing Translation File Click a nominal variable to translate its categories Go to the Edition tab of the ribbon an
30. unassigned record When creating clusters with SQL expressions unassigned records are the observations that cannot be described by the SQL expressions and are left outside the cluster upper bound An upper bound of a subset S of some partially ordered set P lt is an element of P which is greater than or equal to every element of S V variable A variable corresponds to an attribute which describes the observations stored in your database In Infinitelnsight components a variable is defined by Type Storage format Role variable pool The variable pool is a repository where the user stores the description of the frequently used variables It is located in the connector store needs to be associated before does nothing if not associated This call stores all the variable information usually edited by the user standard description storage type value etc mapping information and structure The information stored in the pool is retrieved on the next guessed description The user can also choose to save only the description of a specific variable CUSTOMER SAP Infinitelnsight 7 0 292 2014 SAP AG or an SAP affiliate company All rights reserved Glossary variable type There are four types of variables continuous variables ordinal variables nominal variables textual variables W weight variable A weight variable allows one to assign a relative weight to each of the observations it describes and a
31. 0 75 0 75 1 00 20 0 80 0 80 1 00 21 0 85 0 85 1 00 a 22 0 90 0 90 1 00 33 0 95 0 95 1 00 Random see Wizard _ alidation MAP KxReporto Sheet 4 l Note compatible with Excel 2002 2003 XP and 2007 vV To Open the Current Graph in a New Window Click the B Pin View button The current graph is displayed in a new window CUSTOMER SAP Infinitelnsight 7 0 120 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification For a Nominal Target On the model curve plot different options allow you to visualize Exact profit values for a point for all the displayed curves Thecurves for the different profit types Detected Lift Normalized and Customized For more information on profit types see Available Profit Types on page 46 vV To Display the Exact Profit Values for a Given Point On the screen Model Curves on the plot click a point on one of the curves presented For instance by clicking a point on any one of the curves whose value on the abscissa is 25 the exact profit values appear Profit Detail X Random 0 25 Wizard 1 Validation 0 69797 vV To Select a Profit Type 1 Onthe screen Model Curves beneath the plot click the drop down list associated with the Profit field The list of profit types appears Profit Type Detected 2 Selecta profit type The corresponding profit curves appear For a Continuous large
32. 1 tchangeParameter Parameters VariableSelection StopCriteria FastVariableUpperBoundSelection true tchangeParameter Parameters VariableSelection StopCriteria QualityBar 0 05 tchangeParameter Parameters VariableSelection StopCriteria SelectBestiteration true Gain Chart Settings tchangeParameter Parameters GainChanConfig Leam false tvalidateParameter delete t m bind TransforminProtocol Default 1 t tgetParameter tchangeParameter Parameters Compress true tvalidateParameter delete t m sendMode learn m openNewStore MODEL_SAVE_STORE_TYPE S MODEL_SAVE_STORE_NAME SMODEL_SAVE_STORE_USER MODEL_SAVE_STORE_PWD m saveModel MODEL_SAVE_SPACE MODEL_SAVE_COMMENT MODEL_SAVE_NAME print Model class Census01 has been saved delete m exit 7 Click the Next button to start the generation process Once the script has been generated the menu 191 Using the Model is displayed CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Saving the Model Once a model has been generated you can Save it Saving it preserves all the information that pertains to that model that is the modeling parameters its profit curves and so on VI To Save the Model 1 On the screen Using the Model click the option Save the Current Model The screen Saving the Model appears SAP Infinite Insight Vx y Z class_Census01 File Help x Sav
33. 1 3 0 O 1 0 O CUSTOMER SAP Infinitelnsight 7 0 268 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering The target mean for each cluster that is the percentage of observations belonging to the target category of the target variable contained in each cluster Depending upon the level of information desired you can choose to generate Only the cluster index to which each observation belongs Predicted Value Only option The cluster index and the disjunctive encoding of the cluster indexes Cluster Id Disjunctive Coding option You can also decide to include in the results file all input variables of the application data set Cluster Id Disjunctive Coding copy dataset option The cluster index and the target mean for each cluster Cluster Id Target Mean option For this Scenario You will apply the model to the file CensusOl csv that you used previously to generate the model In the procedure To Apply the Model to a New Data Set Select the format Text files n the Generate field select the option Cluster Id Target Mean Select the folder of your choice in which to save the results file Results generated by the model Analyzing the Results of the Application For this Scenario Open the results text file in Microsoft Excel generated when you applied the model to the CensusOl1 csv file VI To Open the Model Application Results File 1 Depending
34. 209642 H5 grad 9 Married civ Exec mana Husband 9 31 Private 45781 Masters 14 Never marr Prof specia Not in fan 10 42 Private 159449 Bachelors 13 Married civ Exec mana Husband 11 37 Private 280464 S0me colle 10 Married civ Exec mana Husband 12 30 State gov 141297 Bachelors 13 Married civ Prof specia Husband 13 23 Private 122272 Bachelors 13 Never marr Adm clerical Own child 14 32 Private 205019 Assoc acdm 12 Never marr Sales Notin fan 15 40 Private 1217 72 Assocvoc 11 Married civ Craft repair Husband 16 34 Private 2454867 th 8th 4 Married civ Transport Husband 17 25 Sel Fempen 176756 HS grad 9 Never marr Farming fis Qwn child 18 32 Private 186624 H5 grad 9 Never marr Machine o Unmarrie 19 38 Private 28687 11th T Married civ Sales Husband 20 43 Self emp n 292175 Masters 14 Divorced Exec mana LUnmarrie 21 40 Private 193524 Doctorate 16 Married civ Prof specia Husband CUSTOMER SAP Infinitelnsight 7 0 212 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 2 Inthe field First Row Index enter the number of the first row you want to display 3 InthefieldLast Row Index enter the number of the last row you want to display 4 Click the Refresh button to see the selected rows A Comment about Database Keys For data and performance management purposes the data set to be analyzed must cont
35. 289 2014 SAP AG or an SAP affiliate company All rights reserved Glossary Sub sampling Sub sampling means selecting a part of a whole if an event cannot be processed as a whole a limited number of measures have to be taken to represent this event Support The support of a rule is a measure that indicates the number of sessions that verify the rule For instance the number of sessions that contains the itemset A B C and the item D T table of data A table of data is a data set presented in the form of a two dimensional table target key The target key is the expected value of the target target variable A target variable is the variable that you seek to explain or for which you want to predict the values in an application data set It corresponds to your domain specific business issue temporal analytical data set A temporal analytical data set is a special case of analytical data set It is the product of a time stamped population by an analytical record the result of this operation can be seen as a virtual table containing attributes values associated with identifiers in relation with the time stamp In other words a temporal analytical data set contains photos or snapshots of a given list of entities taken at a given time this time can be different for each entity and an entity can be associated with several photos Note Analytical data sets are used to train predictive descriptive models and to appl
36. All rights reserved Infinitelnsight Modeler Regression Classification Understanding the Cost Matrix This section allows you to visualize your profit depending on the selected score or to automatically select the score depending on your profit parameters For each observation category enter a profit or a cost per observation The total profit is automatically displayed on the right of the table To know the threshold that will give you a maximum profit for the profit parameters you have set click the button Maximize Profit Example In the following profit cost table each positive observation correctly identified will yield 15 but each negative observation identified as positive will cost you 8 Category Predicted 1 Predicted 0O True 1 15 O True O 8 O Decision Tree The panel Decision Tree allows you to display the results of a regression or classification model generated by Infinitelnsight Modeler as a decision tree based on the five most contributive variables CUSTOMER SAP Infinitelnsight 7 0 146 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Displaying the Decision Tree vV To Display the Decision Tree for a Target 1 Inthe Target list select the target for which you want to display the decision tree a SAP InfiniteInsight VX y 2 class_Census01 File Help ali Decision Tree Target class E S Save asimag
37. Classification Understanding the Deviations Analysis The first step to take to know if there are any deviations in your data is to loOK at the debriefing report on page 157 and compare the performances KI and KR obtained on the original data with those obtained on the control data set Then to visualize which variables have changed you should loOK into the Control for Deviations Reports on page 157 Debriefing Report The section Control for Deviation Overview provides you with basic statistics on the Data Set used for Deviation Control also called control data set such as the name of the data set Data Set the source file Source the number of records contained in the data set Number of Records and the number of variables for which SAP Infinitelnsight has found deviations in comparison to the data set originally used to train the model Number of variables showing deviation The second and third section of the debriefing report allows you to compare the performance of your model on the original data set with the its performance on the control data set the section Performance Indicators displays for each target the KI and KR indicators obtained by the model on the original data set the section Performance on Control Data Set displays for each target the KI and KR indicators obtained by the model on the control data set If the KI and or KR of the model on the control data set are significantly lower i
38. Copy button and select the desired option The application copies the parameters of the plot You can paste it into a spreadsheet program such as Excel and use it to generate a graph CUSTOMER SAP Infinitelnsight 7 0 119 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification vV To Save the Model Graph 1 Click the kl Save button A dialog box appears allowing you to select the file properties 2 Type a name for your file 3 Select the destination folder 4 ClickOK The plot is saved as a PNG formatted image vV To Print the Model Graph 1 Click the S Print button situated under the title A dialog box appears allowing you to select the printer to use 2 Select the printer to use and set other print properties if need be 3 Click OK The report is printed vV To Export the Model Graph to Microsoft Excel Click the El Export to Excel button situated under the title An Excel sheet opens containing the model graph you are currently viewing along with its data A B C D E F G H l J K 1 Performance percentage 2 ee percentage Random Wizard Validation 4 0 00 0 00 0 00 be 0 05 0 05 0 21 6 0 10 0 10 0 42 i 0 15 0 15 0 63 B 0 20 0 20 0 64 9 0 25 0 25 1 00 10 0 30 0 30 1 00 Performance 11 0 35 0 35 1 00 12 0 40 0 40 1 00 13 0 45 0 45 1 00 14 0 50 0 50 1 00 15 0 55 0 55 1 00 16 0 60 0 60 1 00 17 0 65 0 65 1 00 18 0 70 0 70 1 00 19
39. Data Set In order to accelerate the learn process and to optimize the resulting model you can apply a filter to your data set For this scenario Do not use the filtering option VI To Filter a Data Set 1 Check the option Add a Filter in Data Set 2 Click Next 12 oe number continuous O 0 T3hours perw number continuous eee E T S E a f5dass number nominal o Z O o O o ooo S S Y oOo 18 Kxindex integer continuous 1 Atomica meaa SO VI To Add a Condition 1 Click the button Add a Condition The window Define a Condition opens i Define a Condition on ED 2 Choose a variable in the first list 3 Choose an operator in the second list 4 Indicate a value in the third list Fora variable with number storage type a value For a variable with string storage choose a variable in the list If the list is empty click the button to extract the variable categories 5 Click OK l Note You can edit a condition by double clicking it vV To Add a Logical Conjunction Click the button Add Logic And or the button Add Logic Or l Note You can change a conjunction by double clicking it CUSTOMER SAP Infinitelnsight 7 0 89 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification vV To Change the Order You can change the order of the nodes to accelerate the filtering process by setting the conditions with the highest probability t
40. Database Pre requisites for the In database Apply Mode This optimized scoring mode can be used if all the following conditions are met the apply in data set table view select statement data manipulation and the results data set are tables coming from the same database the model has been computed while at least one physical key variable was defined in SAP Infinitelnsight there is a valid Infinitelnsight Scorer license for the database no error has occurred the in database apply mode is not deactivated granted access to read and write create table To Use the In database Apply Mode Check the option Use the Direct Apply in the Database i SAP InfiniteInsight VX Yy 1 class_Census01 File Help X Applying the Model Application Data Set Data Type DataBase Folder ks wmultiz_sqlserwver2005 w pats Seltieiriem Metadata are stored in the same place as data source Define Mapping Generation Options Generate Predicted Value Only w Advanced Apply Settings Mode Apply w Results Generated by the Model Data Type DataBase Folder Ksrvmulti2_sqlserver2005 al Data model A Define Mapping Use Direct Apply in the Database EJ The inputfileis missing EI EENI I gt Apply Advanced Apply Settings Copy the Weight Variable This option allows you to add to the output file the weight variable if it had been set during the variable selection of the
41. F false positive incorrect assignments to the signal class fluctuation Evolution of the signal that is not stable neither cyclic Infinitelnsight Modeler Time Series G GINI index The GINI statistic is a measure of predictive power based on the Lorenz curve It is proportionate to the area between the random line and the Model curve CUSTOMER SAP Infinitelnsight 7 0 279 2014 SAP AG or an SAP affiliate company All rights reserved Glossary horizon wide MAPE This quality indicator for the forecasting model is the mean of MAPE values observed over all the training horizon A value of zero indicates a perfect model while values above 1 indicate bad quality models A value of 0 09 means that the model takes into account 91 of the signal or in other words the forecasting error model residues is relatively of 9 In database apply The in database apply is used to apply a model into a database the proper SQL code is generated for the model the resulting code is then executed as a single SQL request in the database This avoids extracting the data from the database and speeds up the writing process of the model outputs item A component of an association rule itemset A group or a Set of items is called an itemset iteration An iteration is a single loop through a cycle such as the design prototype test cycle CUSTOMER SAP Infinitelnsight 7 0 280 2014 SAP AG or an SAP affiliate comp
42. Max relationship Husband me sex Male workclass Private qe native country United States race White qe Score of class Proba of class 1 44 Cancel 4 Previous 6 You can modify the value of an explanatory variable and run the simulation again to measure the effect of that change with respect to the target variable For instance 1 Assign the value Widowed to the variable marital status in place of the value Married civ spouse 2 Runthe simulation The probability now obtained is 0 0040 7 Click the Reset button to run the simulation again CUSTOMER SAP Infinitelnsight 7 0 180 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Refining a Model SAP Infinitelnsight allows you to refine a currently open model For instance you can Reduce the number of explanatory variables used by the model while maintaining the initial quality KI and robustness KR Generate a model of degree 2 using the most significant variables of the degree 1 model l Note If your data set contains date or datetime variables automatically generated variables will appear in this panel For more information refer to section Date and Datetime Variables Automatically Generated Variables on page 30 VI To Refine a Model 1 On the screen Using the Model click the option Select Variables The screen Selecting Contributory Variables appears a
43. Metadata Repository Enable Single Metadata Repository Edit Variable Pool Content Graphic Profit Curve Points Bar Count Displayed No Infinitelnsight LoOK and Feel Display 3D Chart Disable Double Buffering Optimize for Remote Display Remember Size and Position when Leaving Report Number of Variables of Interest Active Style Sheet Customize Style Sheets Customizing Style Sheets SAP Infinitelnsight offers the possibility to customize the generated reports The default style sheet called SAP Infinitelnsight Report Style Sheet default cannot be modified You have to create your own style sheets to modify the settings l Note To create load or save a style sheet you have to indicate a data source in the panel Edit Options before opening the window SAP Infinitelnsight Report Style Sheet Editor CUSTOMER SAP Infinitelnsight 7 0 63 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification VI To Create a New Style Sheet In the field Folder click the button J Browse 2 Selecta folder This folder is your style sheets repository 3 Click the button oi Add A new style sheet has been created 4 Click the button id The panel Report Style Sheet Editor opens 5 Inthe field Style Sheet Name enter a name for the new style sheet The extension krs is automatically added l Note You can duplicate a style sheet by cha
44. Proportion above Median 1 Proportion above Median 0 5 1 0 5 0 25 CUSTOMER SAP Infinitelnsight 7 0 245 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Grouping Categories On the plot of details of a variable categories may appear grouped When the option Optimal Grouping is enabled SAP Infinitelnsight groups those categories sharing the same effect on the target variable For instance for the variable education the categories Doctorate and Prof School are grouped If the explanatory variable is continuous SAP Infinitelnsight identifies the points where behavioral changes occur with respect to the target variable and automatically crops the variable into intervals exhibiting homogeneous behavior with respect to the target For more information please see section Optimal Grouping for all Variables r 4 SAP InfiniteInsight Vx y z class_Census01 Variable education Influence on Target 0 125 0 100 0 075 0 050 0 025 oO 000 0 025 O080 0 075 0 100 O 125 0 150 TDoctorate Frot school Ml asters Bachelors Assoc acdm Assoc voc Categories Some college H5 grad 10th 11th 12th 1st Hh 5th 6th 7th 8th 9 BB Yalidation 44 Cancel 4 Previous When categories do not contain sufficient numbers to provide robust information they are grouped in the KxOther category that is created automatically When a v
45. SAP Infinitelnsight IN THIS CHAPTER gi tcele 0 s e 5 Dee ee ane een oh ee ee nen eee ee ee eee Oe ee ee ee een ee eee ee eee 10 Ar hitect re and Operations secisscedec co aceascesscmnecien cceclnavenncteeactasdosccmeenon EE E E date eee ncemedeld aecpiowenaccusdesdeunteupoceoiss 11 Methodological Prerequisites cccccccccssseccceseeeeeeeeeeeeeeeeeeeseeeeeeeseeeeeeeeseaeeeeeesseseeeeeseeeeeeeesegeeeeesssaeeeesssaeeeeeesaaees 15 3 1 Introduction SAP Infinitelnsight platform has be developed in order to provide the ideal Data Mining solution for modeling your data as easily and rapidly as possible while maintaining relevant and readily interpretable results Thanks to SAP Infinitelnsight you will transform your data into knowledge in order to make timely strategic and operational decisions SAP Infinitelnsight places the latest Data Mining techniques within reach of any non expert user SAP Infinitelnsight allows you to access many data source formats and to generate explanatory and predictive models as well as descriptive models in a semi automated manner extremely rapidly With SAP Infinitelnsight you can concentrate on high value added activities such as analysis of the results of data modeling and decision making CUSTOMER SAP Infinitelnsight 7 0 10 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight 3 2 Architecture and Operations The figure below illustrates t
46. SAP Infinitelnsight 7 0 24 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 5 2 Synonyms of Observation and Variable Depending upon your profile and your area of expertise you may be more familiar with other terms that refer to observations in rows and variables in columns when using tables of data The following table presents such terms or synonyms Terms equivalent to Terms equivalent to the term Observation the term Variable Row Column Record Attribute Table Field Event Property Instance Example 4 5 3 Data Formats Whatever the data source used the following two constraints must be accommodated The data must be represented in the form of a single table except in instances where you are using the Infinitelnsight Explorer Event Logging or Infinitelnsight Explorer Sequence Coding features The target variable must be defined for each observation in the table In the sample file CensusOl csv the variable class has been be defined for each individual l Note For information about data formatting and specifically for the list of supported ODBC compatible sources see the document Data Modeling Specification 4 6 Variables 4 6 1Generic Definition A variable corresponds to an attribute which describes the observations stored in your database In SAP Infinitelnsight features a variable is defined by Type Storage format Role CUSTOME
47. Select the desired option 3 Click OK VI To Select the Number of Variables This criterion is mandatory and allows fixing the minimum and the maximum number of variables in the final model 1 Inthe sentence defining the number of variables Se ectthe best model keeping between 1 andall variables select the minimum number of variables for example 1 variable and the maximum number of variables for example all variables 2 For the minimum number of variables a slide is displayed ranging from 1 to the total number of variables in the model Move the cursor on the slide to select the quantity of your choice For the maximum number of variables you can either confirm the minimum number selected previously by choosing Keep all variables or choose a maximum number of variables 3 Click OK Choosing the stopping Criteria You can choose between two variables selection parameters Each step removes variable This option allows you to set the number of variables that should be excluded at each iteration Fach step keeps 95 0 of information CUSTOMER SAP Infinitelnsight 7 0 105 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification This option allows you to set the amount of information that should be kept at each iteration thus limiting the loss of information Select the desired option M amp 3 ol 2 3 M To Select the Number of Var
48. a pem eme emea o o e O o re 3 T number continuous Oe tring E E e ant E e a E E r a S martalstatus sting nominal 0 ______Toccupation sting nomina 0 _____ 8 relationship sting nominal 0 O Z ooo S o O O o o ae string nominat 0 O ooo S y O oOo o Ose string nominal 0 o o o S O aoo Mcapitagain number continuous 0 99999 o O To 12 capitalloss_ number continuous 0 ____13hours perw number continuous 0 O O oo o O O yoo E A EE a a E E e a a o t5dass number nominal fo o o S y O yo E a e a eS _ Add Filter in Data Set 3 Click the option From Statistics in the Extract part A progression bar is displayed while the structure is being extracted Once the extraction is done the icons corresponding to the selected variables change indicating that the operation was a success and allowing you to easily identify them You can then modify the variable structure as you need To Import the Variable Structure from a Model 1 Select the variables for which you want to extract the structure 2 Goto the Structures tab of the ribbon the available options are separated in two parts Edit and Extract 3 Select the option From Model and choose the desired option The window Loading Model opens lt Loading Model i x Data Tyre e Folder c Users denise or tiz caso Documents Models x i Browse Class Version Date 78 K2R Census Class
49. adjoining intervals 2 Click the Remove button The previous and next intervals are extended to include the values previously contained in the deleted intervals so that no gap Is left between intervals structure for an Ordinal Variable The structure for an ordinal variable is similar to that of a continuous variable with the exception of the bounds which are always closed and cannot be modified A Caution The structure for an ordinal string variable cannot be edited CUSTOMER SAP Infinitelnsight 7 0 83 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Structure for a Textual Variable The structure for a textual variable cannot be edited Structure for a Nominal Variable The structure for a nominal variable is made of groups containing the variable categories amp relationship Eq Group Structure Category Edition J Husband Wife Husband a Wife J Not in family i Not in family J Other relative i Other relative J Own child lt Own child J Unmarried Add New Group Add Category i Unmarried New Category Remove Group Remove Category Merge Add Missing __ Alphabetic Sort gt Advanced OK Cancel vV To Create a New Category Group 1 Inthe list Category Edition select the categories you want to add in a new group Use the Ctrl key to select several categories wt Structure Group Structure Catego
50. application solutions 5 1 6 Your Solutions To select the individuals to whom you will send a mailing you have several possible solutions You can use A shotgun method An intuitive method A classical statistical method for example neural networks Bayesian networks logistic models decision trees The Infinitelnsight method CUSTOMER SAP Infinitelnsight 7 0 57 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification shotgun Method This method consists of performing no selection on your database and sending out a mass mailing to every person recorded in your database This solution guarantees that all persons likely to purchase your product are contacted On the other hand this runs to exorbitant costs far exceeding your budget and is seldom the solution applied In addition it runs the risk of saturating the prospects of your bank with inappropriate offers Spamming Intuitive Method This method consists of performing a selection that leans on your knowledge of your field that is to say you send your mailing to individuals selected in an intuitive manner from your database This solution allows you to significantly reduce the cost of your marketing campaign and make it fit your budget This method is not optimal because it does not allow you to Control the real costs and return on investment of your marketing operation Select which
51. be at least one variable set as Order in the Event data source A Caution If the data source is a file and the variable stated as a natural order is not actually ordered an error message will be displayed before model checking or model generation Missing the string used in the data description file to represent missing values e g 999 or Empty without the quotes Group the name of the group to which the variable belongs Variables of a Same group convey a same information and thus are not crossed when the model has an order of complexity over 1 This parameter will be usable in future version Description an additional description label for the variable Structure this option allows you to define your own variable structure which means to define the variables categories grouping CUSTOMER SAP Infinitelnsight 7 0 73 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Viewing the Data To help you validate the description when using the Analyze option you can display the first hundred lines of your data set Vl To View the Data 1 Click the button View Data A new window opens displaying the data set top lines a InfiniteInsight Sample Data View Data Set Census0i csw First Row Index 1 Last Row index 100 a Statistics Graph ai age workclass fntwat education education marital st occupation relations 1 39 Sta
52. belongs to cluster 1 will be displayed in the column kc_dist_cluster_Age_1 V To Add the Probabilities for All Clusters Check the All option CUSTOMER SAP Infinitelnsight 7 0 267 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering vl To Select the Probabilities for Specific Clusters 1 Check the ndividual option 2 Click the gt gt button to display the cluster selection table 3 Check the clusters for which you want to add the probabilities 1 Note When the SQL mode is activated the notion of nearest cluster does not exist If a case belongs to a cluster probability is set to 1 If a case does not belong to a cluster probability is set to O Miscellaneous Outputs Disjunctive Coding This option allows you to add to the output file the disjunctive coding of the clusters A column is generated for each cluster and contains either O or 1 depending whether the observation belongs to the cluster or not The columns created are named kc_disj_ lt TargetVariable gt _ lt Clusterld gt For example if the target variable is Age and the model has five clusters the five following columns will be generated Ac_d s _age_l kc_dis _age_2 kc_dis _age_3 kc_disj_age_4 kc_dis _age_5 Target Mean larget Key Probability This option allows you to add to the output file for continuous targets the mean of the target for the cluster containing the observation displa
53. can define the number of clusters that you want to obtain These fields allow you to specify how many clusters will be generated by the model By default the number of clusters is set to 10 The higher the number of segments the lower the robustness KR The lower the number of segments the lower the information KI One should generally start with the default number and then go further with more or less clusters based on the results For supervised segmentation that is to say with a Target the user chooses the best number of segments for instance 5 10 which means that 5 to 10 clusters are requested by the user The engine computes the best number of clusters using the metric KI KR For instance you may have 7 clusters For unsupervised segmentation that is to say without Target the SAP Infinitelnsight engine chooses the minimum number of clusters for instance 10 10 which means that 10 clusters are requested by the user l Note When you activate the option Calculate SQL Expressions SAP Infinitelnsight generates an additional cluster that contains the unassigned records For more details on SQL expressions and unassigned records see Difference Between Standard Cross Statistics and SQL Expressions on page 260 Choosing to Calculate SQL Expressions allows you to see in the model debriefing the SQL Expressions used to generate each cluster vV For this Scenario Keep the default settings setting Up the Advan
54. combined with a large number of records in the category acombination of both SAP Infinitelnsight uses a non parametric setting in which the category importance is defined as Freg Xi Categorylmportance Xi normalPro fit Xi where normalProfit Xi is the normal profit of category Xi see below for a definition Freq Xi is the global frequency of the category XI Zisanormalization constant We give below the details of the computation of these quantities CUSTOMER SAP Infinitelnsight 7 0 134 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Normal Profit Each category of the target S is associated with a profit profit Sj defined such that j E 2 pro fit S Freq Sj 0 j l The profit of a target category is a value in the range 1 It is defined the following way from the cumulated target category frequencies profit s 29 _ Freq S Freq 5 1 The normal profit of a category Xi is then defined as j B normalProfit X profit S Prob 5 Xi j l Where Proba Sj Xi is the conditional probability of observing the target category Sj in the variable category Xi cross Statistics rede _ Freq 5 X Prob s X Ta The fact that these formulas rely only on frequencies makes them resistant to any monotonic transformation of the target S Normalization Constant The normalization can be approximated
55. complexity of the procedures associated with statistical analysis Using SAP Infinitelnsight you will be able to create a model that allows you to Determine which individuals have the highest probability score of being interested in your marketing campaign predictive modeling You may then apply the model to your entire database Break out the determining factors that describe the phenomenon that you hope to model that is the fact of being interested or not interested in the new financial product of the bank descriptive modeling The profit curve an important validation and control tool allows you to compare the performance of models generated using SAP Infinitelnsight features with that of a hypothetical random model and that of a hypothetical perfect model At the same time it also allows you to determine the optimal number of persons to contact to maximize the profit generated by your campaign SAP Infinitelnsight also provides you with indicators of the quality of the model you generate predictive power and its capacity to generalize and remain relevant to new data sets prediction confidence SAP Infinitelnsight provides you with the means to customize your direct marketing campaign with respect to your different customer profiles increasing your powers of persuasion CUSTOMER SAP Infinitelnsight 7 0 59 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Cl
56. data CUSTOMER SAP Infinitelnsight 7 0 154 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification selecting the Data Set to Analyze First you need to select the data set for which you want to analyze the deviations For the results to make sense the new data set should contain the same columns as the data set that was originally used to train the model including the target variable which must be filled V To Select a Data Set 1 Onthe screen Analyze Deviations select the data source format to be used Text file ODBC 2 Click the Browse button The following selection dialog box appears Select Source Folder for Data d a ae CiUsersinatacha yamiDocuments Ay P Aol HES Gl Census _ BG JapaneseData GG KAR BG KelData eG KSN BG KTC Gets AG cA ee ee Gl KA H PaA a g TA a g T HG ue Samples Text Files dat data csv t ka Use r Password oK Cancel 3 Select the file you want to use then click OK The name of the file appears in the Data Set field 4 Click Next The screen Deviation Analysis Debriefing is displayed CUSTOMER SAP Infinitelnsight 7 0 155 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Following the Deviations Analysis Progress The panel Deviation Analysis Debriefing allows you to follow
57. distributed distribution operations begin again at step 1 b If the entire 4 5 of the initial data set has been distributed distribution operations go to step 4 4 The final 1 5 of the initial data set is sent as a block of data to the test sub set Periodic Without Test The Periodic without test strategy distributes the whole initial data set in a periodic manner to the two sub sets of estimation and validation 3 4 of the initial data set are distributed to the estimation sub set m 1 4 to the initial data set are distributed to the validation sub set In other words this cutting strategy is implemented by following this distribution cycle 1 Three lines of the initial data set are distributed to the estimation sub set 2 Oneline is distributed to the validation sub set 3 Distribution begins again at step 1 As no test sub set is used all the data from your training data set can be used for sub sets of estimation and validation This can lead to a model with a better quality and robustness Sequential The Sequential strategy cuts the initial data set into three blocks corresponding to the usual cutting proportions The lines corresponding to the first 3 5 of the initial data set are distributed as a block to the estimation data set The lines corresponding to the next 1 5 of the initial data set are distributed as a block to the validation data set The lines corresponding to the final 1 5 of the initial data set ar
58. down list select among these options Chessboard maximum of absolute differences between coordinates LInf Euclidean square root of sum of square differences between coordinates L2 City Block sum of absolute differences between coordinates L1 System Determined default value Lets the system determine the best distance to be used according to the model build settings l Note The current policy is to use LInf either in unsupervised mode or when the clusters SQL expressions have been asked for and L2 otherwise Choosing the Encoding Strategy The Encoding Strategy option refers to the kind of encoding the segmentation engine is expecting from the data encoder of Infinitelnsight vV To Choose an Encoding Strategy Choose among the following options from the drop down list CUSTOMER SAP Infinitelnsight 7 0 224 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 225 Option System Determined Target Mean Uniform Unsupervised Description Lets the system select the best encoding according to the model parameters The Target Mean encoding is used for supervised models Otherwise variables are encoded using the Unsupervised scheme Default value for supervised clustering Each value of a continuous input variable is replaced by the mean of the target for the segment the value belongs to Each category of a nominal input variable is
59. during the installation process you will find the Samples folder located in C Program Files SAP Infinite nsight InfinitelnsightVx y Z 4 Select the file CensusOl csv then click OK The name of the file appears in the Estimation field 5 Click Next The screen Data Description appears 4 SAP InfiniteInsight VX y 2 New Regression Classification Model File Help Data Description iB Open Description Save in Variable Pool a i Analyze Save Description Remove from Variable Pool View Data Properties Description Index Name Storage Value Key order Missing Group TDescription Siruchi re _ Add Filter in Data Set 44 Cancel Eg You must analyze the data or open a description file Al Previous Ib Next 6 Goto section Describing the Data Selected CUSTOMER SAP Infinitelnsight 7 0 209 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Describing the Data Selected 210 For this scenario Select Text Files as the file type Use the file Desc_CensusOl csv as the description file for the CensusOl cvs data file To Select a Description File On the screen Data Description click the button Open Description The following window opens i Load a Description for Census01 csv Folder SBamplesiCensus 1 Browse File Name o In the window Load a Description select the type of your description file In the Folde
60. encoding based on the distribution curve to map the target values into the range 1 1 The curve is different for each sub data set You can access to this curve through the parameter tree under the section UniformCurvePoints 2 Sortthe normalized target values and graph the cumulative sums to obtain the Wizard graph Note that 20 bins are used to give a good approximation while using less computation 3 Re sort by estimated values and again graph the corresponding actual values cumulative distributions validation graph 4 As usual the predictive power is the ratio of the validation and wizard areas The predictive power is thus based on the order of the estimates and compares this order with the order of the actual continuous targets As such it is more robust than the L1 such as Mean absolute Error MAE or L2 metrics such as Mean square error MSE or Root Means Square Error RMSE or Pearson coefficient often used for regression since one very large erroneous target will never decrease the overall predictive power figure and could be a major cause for instability of all the other metrics On the other hand the predictive power does not take into account the estimated values with respect to the target values in other words a model with estimates in the range of 2 2 could have a very good predictive power even if the actual targets are in the range 0 100 provided that the model has found the correct order between estimat
61. hours per week in J NOT El AND 0 81 capital gain in KxMissing or 4386 41310 F AND 4 74 6 2 be education in Bachelors Masters Profschool Doctorate Marital status in Never married Married spouse absent Divorced Separated Widowed H AND 17 31 F AND 3 56 The SQL expression can be brOKen down as follow the first part 1 defines a cluster of observations where the variables equal the values displayed the second part 2 defines clusters of observations that are excluded for the cluster found in part 1 The percentages displayed indicate the proportion of each cluster excluded with respect to the cluster found in part 1 In our example the first excluded cluster corresponds to observations where the capital gain variable has its value ranges between 4386 excluded and 41310 4386 415107 It represents 0 81 of the observations found in part 1 l Note The clusters are created by applying the SQL expressions in a specific order defined by the engine If you apply the SQL rules randomly you may not obtain exactly the same result CUSTOMER SAP Infinitelnsight 7 0 259 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Difference Between Standard Cross Statistics and SQL Expressions When you ask for SQL expressions the final segmentation is different from the one without The goal of SQL is to have easy to understand and easy t
62. identifies the points where behavioral changes occur with respect to the target variable and automatically crops the variable into intervals exhibiting homogeneous behavior with respect to the target For more information please see section Optimal Grouping for all Variables Fa SAP InfiniteInsight Vx y 2 class_Census01 File Help al Category Significance Data Sets View Type Copy Print Save Export to T Orientation x Excel E OF sB A Bar Pin View Variables educa tion Variable education Influence on Target 0 125 0 100 0 075 0 050 0 025 oO000 0 025 0o50 0 075 0 100 0 125 0 150 Doctorate Prof school hl asters Bachelors Assoc acdm Assoc voc Categories Same college H5 grad 10th 11th 12th 1st 4th Sth 6th 7th 8th g Bi Validatian 4i Cancel CUSTOMER SAP Infinitelnsight 7 0 136 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification When categories do not contain sufficient numbers to provide robust information they are grouped in the KxOther category that is created automatically When a variable is associated with too many missing values the missing values are grouped in the KxMissing category that is also created automatically To understand the value of the categories KxOther and KxMissing consider the following example The database of corporate customers of a business contains the v
63. integer date datetime Value the value of the constant date format yyYyY MM DD datetime format YYYY MM DD HH MM SS Key indicates if the constant is a key variable or Q the variable is not an identifier identifier for the record You can declare multiple keys They will be built according to 2 primary identifier the indicated order 1 2 3 2 secondary identifier vV To Define a Constant Click the Add button A pop up window opens allowing you to set the constant parameters In the field Output Name enter the constant name In the list Output Storage select the constant type In the field Output Value enter the constant value a Ff O N a Click OK to create the constant The new constant appears in the list You can choose whether to generate the defined constants or not by checking the Visibility box CUSTOMER SAP Infinitelnsight 7 0 265 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Outputs by Cluster Identifier Top Ranking Centroids Indices This option allows you to add to the output file the number of the clusters whose centroids are the closest to the current observation The closest cluster is the one the observation belongs to its number is displayed in the column kc_ lt Target variable gt The next closest cluster is displayed in the column kc_ lt Target Variable gt _2 and so on until the furthest cluster you can choose to add all the c
64. interval any value higher than its higher bound vl To Create a New Interval 1 Click the Add button to create a new interval The edit window opens x Include Smaller Data Include Higher Data Interval 0 Minimum Maximum 0 Add Missing age xi Q al fe Add Missing E Split Merge Add Remove gt Advanced OK Cancel Select the lower bound type by clicking the J button Enter the minimum value for the interval in the left text field Enter the maximum value for the interval in the right text field Select the higher bound type by clicking the button Check the option Add Missing if the missing values must be grouped with this interval O oo fF Q N age x g a fo fea id Add Missing CUSTOMER SAP Infinitelnsight 7 0 82 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 7 Click the Yes button to validate your interval vV To Split an Interval 1 Select the interval to split 2 Click the Split button The selected interval is automatically split into two equal intervals vV To Merge Two Intervals 1 Select the intervals to merge You can only select adjoining intervals 2 Click the Merge button _ Include Smaller Data Include Higher Data Interval D Minimum maximum o Ada missing umim gt Advanced OK Cancel vV To Delete an Interval 1 Select one or more intervals You can only select
65. option with the Decision option described below you can link the best score with the category that has obtained it CUSTOMER SAP Infinitelnsight 7 0 170 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Decision This option allows you to generate in the output file the best decision s for each observation Like for the previous option the scores obtained for each category of the target variable are compared and the category with the best score for the current record is displayed in the column decision_rr_ lt Target Variable gt if several decisions have been requested the category with the second best score Is displayed in the column decision_rr_ lt Target Variable gt _2 the one with the third best score in the column decision_rr_ lt Target Variable gt _3 and so on Probabilities This option allows you to generate in the output file the probability of the best decisions for each observation Like for the previous options the scores obtained for each category of the target variable are compared and the probability of the category with the best score for the current record Is displayed in the column proba_rr_ lt Target Variable gt If several decisions have been requested the probability of the category with the second best score is displayed in the column proba_rr_ lt Target Variable gt _2 the one with the third best score in the column proba_rr_ lt Target Variable gt
66. or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Vv To Print the Model Overview La 1 Click the Print button situated under the title A dialog box appears allowing you to select the printer to use 2 Select the printer to use and set other print properties if need be 3 Glick OK The report is printed VI To Save the Model Overview Click the k Save button situated under the title The file is saved in HTML format vV To Export to PowerPoint PPT Export to Click the PowerPoint Export to PowerPoint button Model Graphs Definition The model graphs allow you to View the realizable profit that pertains to your business issue using the model generated Compare the performance of the model generated with that of a random type model and that of a hypothetical perfect model On the plot for each type of model the curves represent the realizable profit Y axis or ordinate as a function of the ratio of the observations correctly selected as targets relative to the entire initial data set X axis or abscissa CUSTOMER SAP Infinitelnsight 7 0 235 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Displaying the Model Graphs vV To Display the Model Graphs 1 Onthe screen Using the Model click the Model Graphs option The model graph appears Performance fan t Lm b i a
67. quantiles decile vingtile percentile V To Compute the Gain Chart 1 Check the box Compute Gain Chart on Apply in Data 2 Inthe list select the Number of Quantiles you want your data to be segmented in 3 You can add additional variables in order to estimate profits per segments of the population 1 Inthe Variables list select the variables you want to add to the gain chart Use the C7AL key to select multiple variables 2 Click the gt button to add the selected variables to the list Values for Gain Chart CUSTOMER 166 SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 4 The sum ofeach selected variable will be calculated for each segment of the population 5 Click Validate to save the advanced parameters and go back to the panel Applying the Model vi Results The result of the gain chart computation is available at the end of the model application It can also be found in the Statistical Reports in the section Model Performance results The result of the gain chart computation is available at the end of the model application a SAP InfiniteInsight VX y 2 class_Census01 File Help A id Fa B C View Sort Series Copy Print Export T w hd w F FA Target ess o Ey Summary Quan Weighted Cou Min Score MaxScore Actual Predicted tie aa ee ee 1 ee css Moo RR 1 009 RR 1
68. replaced by the mean of the target for this category In case of a nominal target variable the mean of the target corresponds to the percentage of positive cases of the target variable for the input variable category Each variable segment is encoded in the range 1 1 so that the distribution of the variables is uniform Default value for unsupervised clustering A target free strategy Only segment frequency is used to encode variables The following options will only be displayed when all variables are continuous Option Natural Min Max Standard Deviation Normalization Description This option does not transform the input data This option encodes the categories of the variable in the range 0 1 where 0 corresponds to the minimum value of the variable and 1 corresponds to the maximum value This option performs a normalization based on the variable mean and standard deviation x Mean Sta Dev Activating the Autosave Option The panel Model Autosave allows you to activate the option that will automatically save the model at the end of the generation process and to set the parameters needed when saving the model vV To Activate the Autosave Option In the panel Summary of Modeling Parameters click the Autosave button Compute Decision Tree __ Enable Auto selection The panel Model Autosave is displayed CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights r
69. spouse absent Categories Separated Widowed Never married This plot presents the effect of the categories of the marita status variable on the target variable CUSTOMER SAP Infinitelnsight 7 0 243 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Variable Categories and Profit The plot Category Significance illustrates the relative significance of the different categories of a given variable with respect to the target variable On this type of plot The higher on the screen one finds a category the greater the positive effect on the target category or hoped for value of the target variable In other words the higher a category appears on the screen the more representative that category is of the target category of the target variable The width and direction of the bar correspond to the profit contributed by that category In other words they correspond to the relationship of that category to the target variable and whether that category has more or less observations belonging to the target category of the target variable For a given category a positive bar on the right of 0 0 indicates that the category contains more observations belonging to the target category of the target variable than the mean calculated on the entire data set A negative bar on the left of O O indicates that the category contains a lower concentration of target categor
70. the analysis process thanks to a progression bar i SAP InfiniteInsight VX Y 2 class_Census01 File Help Analyze Deviations Debriefing i BLA B w oH Pee Stop ViewType Copy Print Save a Checking Deviations a Computing additional performances Stop Current Task 44 Cancel dl Previous I gt Next At the end of the process a debriefing panel is displayed For details on the debriefing panel see section Understanding the Deviation Analysis on page 157 You can use the toolbar provided on the upper part of the panel to dea stop the analysis process by clicking the ha a button r display the text log detailing the process by clicking the button copy print or save the debriefing panel vV To Copy the Report Click the Copy button The application copies the HTML code of the screen You can paste into a word processing or spreadsheet program a text editor vV To Print the Report YS 1 Click the Ta Print button situated under the title A dialog box appears allowing you to select the printer to use 2 Select the printer to use and set other print properties if need be 3 Click OK The report is printed vV To Save the Report Click the k Save button situated under the title The file is saved in HTML format CUSTOMER SAP Infinitelnsight 7 0 156 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression
71. the group and re added to the list Category Edition Working Without any Defined Structure If you let the structure as undefined SAP Infinitelnsight using consistent coder automatically determines the categories grouping depending on their interaction with the target variable You can configure two parameters in this case 86 the band count for continuous variables Infinitelnsight Modeler Data Encoding optimal grouping for all variables CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Band Count for Continuous Variables When you work with no defined structure you can set the band count for continuous variables The allowed values for this parameter are between 1 and 20 The population is thus divided into as many segments of similar size These segments are used to build descriptive statistics particularly the distribution of target variables for each segment which affects the coding of the variable with respect to target variables The band count has an influence on the calculation of the predictive power the more there are segments the more accurate is the calculation of the predictive power for the explanatory variable However this influence is very small V To Edit the Band Count for Continuous Variables 1 Click the row corresponding to the continuous variable to be edited 2 Goto the Structures tab of the r
72. the name of the user defined constant Storage the constant type number string integer date Value the value of the constant Key indicates if the constant is a key variable or identifier for the record You can declare multiple keys They will be built according to the indicated order 1 2 3 Vl To Define a Constant oa Ff WS N a Value Warnings checked the constant appears in the output unchecked the constant does not appear in the output 1 The name cannot be the same as the name of an existing variable of the reference data set 2 Ifthe name is the same as an already existing user defined constant the new constant will replace the previous one number string integer date datetime date format YYYY MM DD datetime format yyYY MM DD HH MM SS O the variable is not an identifier 1 primary identifier 2 secondary identifier Click the Add button A pop up window opens allowing you to set the constant parameters In the field Output Name enter the constant name In the list Output Storage select the constant type In the field Output Value enter the constant value Click OK to create the constant The new constant appears in the list You can choose whether to generate the defined constants or not by checking the Visibility box Gain Chart This tab allows you to compute the gain chart on the apply data set that is to rank your data in order of descending scores and split it into exact
73. the variable ranges for a specific cluster are displayed click the black horizontal bar a moving cursor is displayed Drag the cursor down to display the list of clusters 3 Above the table from the drop down list associated with the Variable field select the variable for which you want to see the profile The cross statistics will appear in the form of a plot in the lower part of the screen CUSTOMER SAP Infinitelnsight 7 0 253 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Understanding Clusters Profiles The screen Cluster Profiles can be brOKen down into three parts Inthe upper part a drop down list allows you to select the variable for which you want to see the cross Statistics Variables are presented in descending order of the significance of their contribution relative to the target category of the target variable When a cluster is selected the variables visible in the drop down list are sorted according to the difference between their cluster profile and their population profile the Kullback Leibler divergence is used to measure this difference The variable that appears first on the list is the variable exhibiting the greatest difference between its two profiles This sorted list of variables provides the set of discriminatory variables required to describe a cluster n the middle part a table presents each cluster in a Summarized fashion Column
74. the variables the target variables the explanatory variables and possibly a weight variable Check the modeling parameters Setting the Advanced Parameters on page 99 degree target key rule mode variable auto selection and correlations This step is optional oF GW N CUSTOMER SAP Infinitelnsight 7 0 66 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification selecting a Data Source For this Scenario Use the file CensusOl csv as a training data set This file represents the sample that you had extracted from your database and used for the test phase of your direct marketing campaign As specified in your test plan this file contains data concerning 50 000 prospects for whom you now know the behavior with respect to the new financial product 25 of the prospects showed themselves to be clearly interested They chose to accept a an invitation for a meeting with one of your sales channel agents 75 of the prospects declined your invitation In this file you created a new variable Class which corresponds to the reaction of prospects contacted during the test You assigned The value 1 to those prospects who responded positively to your invitation The value O to those prospects who responded negatively to your invitation VI To Select a Data Source 1 On the screen Select a Data Source select the data source format to be used Text files Data Ba
75. this case decision_rr_class The probability decision is also based on the score and provides the probability of the decision The higher it is the more it will confirm the decision value The name of this column corresponds to the name of the target variable prefixed by proba_decision_rr_ or in this case proba_decision_rr_class The probability for each observation that it does or does not belong to the target category of the target variable The name of this column corresponds to the name of the target variable prefixed by proba_rr_ or in this case proba_rr_class The prediction range or maximum error The name of this column corresponds to the name of the target variable prefixed by bar_rr_ or in this case bar_rr_Class The individual contributions by variables contained in the data set with respect to the target variable The names of the columns of individual contributions correspond to the names of each of the variables prefixed by contrib_ or in this case contrib_age contrib_workclass and so on CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Performing a Simulation The open model may be used to carry out simulations on specific observations one at a time To define the observation to be analyzed the variables of your choice must be associated with values For instance if you have selected the occupatio
76. to make it available to other SAP Infinitelnsight features Infinitelnsight Explorer Sequence Coding formerly known as KSC aggregates events into a series of transitions For example a customer click stream from a Web site can be transformed into a series of data for each session Each column represents a specific transition from one page to another Similar to Infinitelnsight Explorer Event Logging these new columns of data can be added to existing customer data and are made available to other SAP Infinitelnsight features for further processing Infinitelnsight Modeler Data Encoding formerly known as K2C automatically prepares and transforms data into a format suitable for use in the SAP Infinitelnsight Infinitelnsight Modeler Data Encoding translates nominal and ordinal variables automatically fills in missing values and detects out of range data In addition this feature contributes significantly to the robustness of the models generated by the SAP Infinitelnsight engine by providing a robust data encoding CUSTOMER SAP Infinitelnsight 7 0 13 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight Phase 3 Data Modeling Thanks to the statistical techniques and information technologies upon which the Infinitelnsight Modeler Regression Classification Infinitelnsight Modeler Segmentation Clustering and Infinitelnsight Modeler Time Series features were built these f
77. using the SAP Infinitelnsight you should State a business issue that you want to solve Possess a data set representing this issue in the form of a set of observations 3 3 1 What is your Business Issue All of the SAP Infinitelnsight features are a response to the same requirement they allow supervised data analysis The term supervised means that the data analysis does not occur completely independently but always as a function of a particular issue your business issue Consider the database that contains information about your customers An analysis that grouped your customers into homogeneous groups independently of your input would be of little interest On the other hand an analysis that grouped them as a function of a variable such as mean business revenues earned from this customer each year would offer significant interest You would learn the characteristic profiles of the customers that bring you the most money Then you can develop strategies to better influence your customers according to their characteristic profiles To recap the prerequisite step before using SAP Infinitelnsight consists of identifying and formulating your business issue 3 3 2 Is your Data Usable Once your business issue has been identified and formulated you need to have data on hand that will permit an answer to be found We will not expound at length about the information value associated with data This depends on your data co
78. variable SAP Infinitelnsight feature generated KxIndex automatically Retain all the other variables To Exclude some Variables from Data Analysis 1 Onthe screen Selecting Variables in the section Explanatory Variables Selected left hand side select the variable to be excluded a SAP InfiniteInsight Vx y z New Regression Classification Model File Help Selecting Variables Explanatory Variables Selected 14 Target Variables 1 workclass fniwat education education num marital status occupation relationship race SEX _ Alphabetic Sort capital gain Weight Variable 0 capitalloss hours per week native country Excluded Variables 1 Al _ Alphabetic Sort H M Alphabetic Sort 1 Note On the screen Selecting Variables variables are presented in the same order as that in which they appear in the table of data To sort them alphabetically select the option A ohabetic sort presented beneath each of the variables list CUSTOMER SAP Infinitelnsight 7 0 221 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Click the button gt located on the left of the screen section Variables excluded lower right hand side The variable moves to the screen section Variables excluded Also select a variable in the screen section Variables excluded and click the button lt to move the variables back to the screen section Explanatory variable
79. w2 X2 wn Xn WLIXI XL wl2X1lX2 F WwWl3Xl X3 sxs F Wi Wy Sha Methodology In the large majority of cases a first degree polynomial is sufficient for generation of a relevant and robust model Using a higher degree of polynomial does not always guarantee better results than those obtained with a first degree polynomial In addition the higher the degree of polynomial you select The more time needed to generate the corresponding model The more time needed to apply the model to new data sets The harder it is to interpret the results of modeling The selection of the degree of the polynomial depends on the nature of the data to be analyzed The recommended method is to First generate a model with a first order model In the large majority of cases this degree will be sufficient to guarantee a relevant and robust model Test the results thus obtained with models of greater degree if the performance of the first order model seems inadequate 4 7 6 Validating the Model Once the model has been generated you must verify its validity by examining the performance indicators The predictive power allows you to evaluate the explanatory power of the model that is its capacity to explain the target variable when applied to the training data set A perfect model possesses a predictive power of 1 anda completely random model possesses a predictive power of O The prediction confidence defines the degree of robustne
80. way risk score fitting is performed that is how Infinitelnsight fits its own scores to the risk scores The risk fitting has two modes PDO based the area equals Median Score N PDO Median Score N PDO N number of PDOs around the median score must be specified by the user By default N is set to 2 l Note PDO stands for Points to double the odds CUSTOMER SAP Infinitelnsight 7 0 108 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Frequency based the area equals Quantile Freq Quantile 1 0 Freq The frequency of higher and lower scores to be skipped must be specified by the user By default the frequency is set to 15 If you do not check the box Risk Fitting Domain the mode Frequency based will be used by default The fitting can be weighted or not vV To Set the Risk Fitting Parameters 1 Check the box Risk Fitting Domain General Sherpas Set SELaCrt m Gain Chart ai Enable Risk Score for good bad odds ratioof 9 to1 Pointstodoubleodds 3B View Score Table Based on paints to double odds Number of points to double odds around median score 2 a Frequency Based Percentage of lower and higher scores to skip 15 Use Score Bin Frequency as Weights Ik OK 2 Select the mode you want to use 3 Depending on the selected mode set the appropriate value in the corresponding field 4 Ifyou want to use weight
81. 0 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Target Variable Definition A target variable is the variable that you seek to explain or for which you want to predict the values in an application data set It corresponds to your domain specific business issue When the target variable is a binary variable the SAP Infinitelnsight considers that the target value or target category of this variable that is the value that is the object of the analysis to be the least frequently occurring value in the training data set Imagine that a training data set containing the customer information of a company contains the target variable responded to my mailing This target variable may take the values Yes or No If the value Yes is the least frequent value for instance if 40 of referenced customers responded to the mailing the SAP Infinitelnsight considers that value to be the target category of the target variable synonyms Depending upon your profile and your area of expertise you may be more familiar with one of the following terms to refer to target variables Variables to be explained Dependent variables Output variables These terms are synonyms CUSTOMER SAP Infinitelnsight 7 0 31 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Example Your company is marketing two products A and B You have a database which contains references
82. 014 SAP AG or an SAP affiliate company All rights reserved Glossary entity An entity is the object of interest of any analytical task it can be a customer a product a store and is usually identified with a single identifier that can be used throughout the data repositories Entities are usually associated with a state model describing the life cycle of such an analytical object of interest Note This is a technical constraint entities MUST be uniquely identified error bar see prediction range error mean mean of the difference between predictions and actual values error standard deviation dispersion of errors around the actual result event data set An event data set should consist of at least an event date such as birthdate or beginning of trial in YYYY7 MM DD format areference id i e customer id that will be used to join the Events or transactions data with the reference or static customer table previously defined excluded variable actual target CUSTOMER SAP Infinitelnsight 7 0 278 2014 SAP AG or an SAP affiliate company All rights reserved Glossary explanatory variable An explanatory variable is a variable that describes your data and which serves to explain a target variable Expression Editor Panel allowing to create fields as complex expressions in the Analytical Data set Editor extra predictable variable Variable whose values are known for the period that is to be predicted
83. 133373 H3S grad 9 Married civ Transport Husband 13 25 Private 189775 Some colle 10 Married sp Adm clerical OQwn child 14 25 Private 181054 Bachelors 13 Never marr Sales Notin fan 15 25 Private 267044 Some colle 10 Never marr Adm clerical Mot in fan 16 25 Private 200408 Some calle 10 Never marr Tech supp Not in fan iT 25 2024860 Assoc acdm 12 Never marr Other rela 18 25 Private 332702 Assoc voc 11 Never marr Other seri Own child 19 25 Private 316688 HS grad 9 Never marr Machine o Not in fan 20 25 Private 499233 H3S grad 9 Divorced Adm clerical Not in fan 21 25 Private 98155 Some colle 10 Never marr Transpor Unmarrie Help CUSTOMER Close SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification vV To Save a Filter You can save the filter you created to be able to reuse it at a later moment without being obliged to recreate the same conditions 1 Click the button Save Filter A pop up window is displayed 2 Inthe list Data Type select the format in which you want to save the filter 3 Use the Browse button located on the right of the Folder field to select the folder or database where you want to save the filter 4 Inthe Description field enter the name of the file or table in which you want to save the filter 5 ClickOK vV To Load an Existing Filter To apply a f
84. 2 2003 XP and 2007 vV To Open the Current Graph in a New Window Click the B Pin View button The current graph is displayed in a new window Understanding the Model Graphs The following figure represents the model graph produced using the default parameters Profit Type Detected v Models tr_class Vv Performance 0 9 0 8 07 4 0 6 5 0 5 4 Detected Profit 0 4 4 0 3 4 0 1 4 0 0 oh ph oh er no ane ap ap or oP er seh aor ep 40 q6 Foia on op op x op La percentage E Random E Wizard Validation CUSTOMER SAP Infinitelnsight 7 0 238 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering On the plot the curves for each type of model represent the profit that may be realized Y axis that is the percentage of observations that belong to the target variable in relation to the number of observations selected from the entire initial data set X axis On the X axis the observations are sorted in terms of decreasing score that is the decreasing probability that they belong to the target category of the target variable In the application scenario the model curves represent the ratio of prospects likely to respond in a positive manner to your marketing campaign relative to the entire set of prospects contained in your database Detected profit is the default setting for type of profit Using this type of profit The val
85. 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 4 InfiniteInsight Sample Data View Data Set age workclass fniwat Census01 csv First Row Index 1 Last Row index 100 E Statistics Graph i education education marital st occupation relations 1 25 Self emp n 176756 HS grad 9 Never marr Farming fis Qwn child 2 25 Private 269980 HS grad 9 Never manrr Handlers c Mot in far 3 25 Private 32275 Some colle 10 Married civ Exec mana Wife 4 25 200681 Some colle 10 Never marr Own child 5 25 Private 252752 H5 grad 9 Never marr Other servi Unmarrie 6 25 Private 255004 10th 6 Never marr Crafi repair Not in fan T 25 Private 159732 Some colle 10 Never marr Adm clerical Mot in fan 5 25 Private 193787 Some colle 10 Never marr Tech supp Qwn child 9 25 Private 371987 Bachelors 13 Never marr Exec mana Notin fan 10 25 Private 344991 Some colle 10 Married civ Craft repair Husband 11 25 Private 86872 Bachelors 13 Married civ Exec mana Husband 12 25 Private 133373 HS5 grad 9 Married civ Transport Husband 13 25 Private 189775 Some colle 10 Married sp Adm clerical Own child 14 25 Private 181054 Bachelors 13 Never marr Sales Not in fan 15 25 Private 267044 Some colle 10 Never marr Adm clerical Not in fan 16 25 Private 200408 Some calle 10 Never marr
86. 5 0 200 31 35 28 30 Categories Ta ais 27 E3 24 19 23 17 19 E Estimation Validation 44 Cancel CUSTOMER SAP Infinitelnsight 7 0 130 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification ae 2 Click Data Sets and select the 2 Validation Only button to go back to the Validation Data Set plot VI To Switch between Curve and Bar Charts 1 Click View Type and select the aj button to display the curve chart The curve plot appears a SAP InfiniteInsight VX y 2 cass_Census01 File Help ali Category Significance A W 0 abb A Data Sets View Type Bar Copy Print Save pasar to pin View i K Sa es Orientation r Variables age ha Variable age Detected Profit at g gD gD gD gD gho gho go oah ah abh ah gtk ah th M a TN gh ah 7 7 7 ri 7 hy sph ph AO gh ae ah a a wt gh at ah gt gt ge e Categories E Fandom E Wizard Estimation F Validation ge ge yar wt gts 2 Click View Type and select the aly button to go back to the bar chart l Note You can combine the different types of plot For example you can display All Datasets in a curve chart or the Validation Data Set in a bar chart CUSTOMER SAP Infinitelnsight 7 0 131 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Understanding the Plots of Var
87. 72 corresponding to an entity see page 2 7 An analytical record may be decomposed into domains see page 277 that group attributes related to each other for example in CRM an analytical record can have a demographic domain and a behavioral domain antecedent X is called the antecedent of the rule The antecedent can be composed of an item see page 280 or an itemset see page 280 for example X can be the set A B C CUSTOMER SAP Infinitelnsight 7 0 271 2014 SAP AG or an SAP affiliate company All rights reserved Glossary application data set An application data set see page 276 is a data set to which you apply a model This data set contains an unknown target variable see page 290 for which you want to know the value association rule An association rule is an implication relation of the form X gt Y The rule means if the attribute X is present in a session then the attribute Y is present too Two measures allow qualifying the quality of the rule the Support see page 290 and the Confidence see page 274 attribute In computing an attribute is a specification that defines a property of an object element or file AUC The AUC statistic is a rank based measure of model performance or predictive power calculated as the area under the Receiver Operating Characteristic curve see ROC on page 47 For a simple scoring model with a binary target this represents the observed probability of a signal respond
88. 90 BR 121 22 Detailed Statistics on Control Grou m a ae i oo C E E T ees TF 3 E e Boos Boz B o ioa 4 E a poos foise 346 1269 46 5 ccs oozes 0 099 Biss fl 724 031 6 sss cos ooze 224 328 193 7 sss Bo 0067 166 153 409 alae D cscs Bos ois 54 29 376 m 9 ccs oa Doos 11 16 767 m 10 M soe Boa 7 3 598 cs Oy It can also be found in the Statistical Reports in the section Model Performance CUSTOMER SAP Infinitelnsight 7 0 167 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Outputs by Targets Reason Codes This feature allows retrieving a list of the variables whose values influence the most a score based decision typically a risk score An example of use of the reason codes is to provide a customer with the reasons why the automatic scoring system did not approve their loan vV To Generate Reason Codes In the tree Advanced Apply Settings located on the left of the panel open the node Outputs for Target Number of reason codes you want to generate Threshold used for computing the most important reason codes For each variable the contribution corresponding to the customer score is compared to its contribution for the whole population The variables for which the contribution is the most differential are selected as the most important reason codes For example if you select Mean the customer variable contribution will
89. AP Infinitelnsight interface simply by selecting the format of the file to analyze Once you have built your model with SAP Infinitelnsight you can generate a SAS table containing the model application results scores probability cluster number predicted value The SAP Infinitelnsight interface allows you to select the output format The generated SAS table is automatically integrated in SAS information system Phase 2 Data Manipulation and Preparation The Infinitelnsight Explorer Sequence Coding and Infinitelnsight Explorer Event Logging features are data manipulation and preparation features They are used to encode data in a robust and semi automatic manner making them available for use by all analytical features of the SAP Infinitelnsight The use of these features is transparent for the final user all data processing is performed in a completely automatic manner Infinitelnsight Explorer Event Logging formerly known as KEL aggregates events into periods of time It allows integrating transactional data with demographic customer data It is used in cases when the raw data contains static information such as age gender or profession of an individual and dynamic variables such as spending patterns or credit card transactions Data is automatically aggregated within user defined periods without programming SQL or changing database schema Infinitelnsight Explorer Event Logging combines and compresses this data
90. Advanced Apply Settings Mode Apply Results Generated by the Model Data Type Text Files W Folder Samples Census i Browse Data E Browse Define Mapping E The input file is missing 41 Previous I Apply CUSTOMER SAP Infinitelnsight 7 0 160 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 161 In the section Application Data Set select the format of the data source in the list Data Type Click the Browse button to select e In the Folder field the folder which contains your data set e In the Data field the name of the file corresponding to your data set In the section Results generated by the model select the file format for the output file in the list Data Type You may also opt to select Keep only outliers If you select this option only the outlier observations will be presented in the results file obtained after applying a model Click the Apply button The screen Applying the Model appears Once application of the model has been completed the results files of the application is automatically saved in the location that you had defined from the screen Applying the Model i SAP InfiniteInsight Vx y 7 class_Census01 File Help Applying the Model a E P amp B Stop View Type Copy Print Save Applying the Model 3 Stop Current Task 44 Cancel qi Previous IP Next CUSTOMER SA
91. Clustering feature of SAP Infinitelnsight respectively These two chapters are organized in the same manner in two parts The first part introduces a detailed application scenario of that feature The second part introduces the actual use of the feature itself using the illustration of the corresponding application scenario A summary and detailed table of contents located at the beginning of the guide an index located at the end and cross references throughout the document allow you to find the information that you need quickly and easily CUSTOMER SAP Infinitelnsight 7 0 4 2014 SAP AG or an SAP affiliate company All rights reserved How to Use this Document 1 2 Which Sections should you Read Depending on your job profile and your needs you may choose to read the entire guide or only certain sections In either case it is essential that you read the section concerning the SAP Infinitelnsight performance indicators on page 39 These indicators embody one of the most important concepts of SAP Infinitelnsight they allow evaluation of the quality and robustness of the models generated The following table provides some points of reference to facilitate your use of this guide What is your Profile You want to evaluate SAP Infinitel nsight and your time is tightly budgeted You want to be guided step by step through SAP Infinitel nsight You have had only limited hands on experience in data modeling
92. Conventions Used in this DOCUMENL ccccccseeseeccceeeeeueeeeeeeeeeeeeaeeeeceeeeesseeeeeeeeeeeeseeeeeeeeeeeesseeesseeeeessaeaaaeeeeeeeeeaaas 6 1 1 Organization of this Document This document is subdivided into six chapters This chapter Welcome to this Guide serves as an introduction to the rest of the guide This is where you will find information pertaining to the reading of this guide and information that will allow you to contact us Chapter 2 SAP Infinitelnsight provides an overview of SAP Infinitelnsight its architecture and its operation In addition it presents two indispensable prerequisite methodologies for using SAP Infinitelnsight features Chapter 3 Essential Concepts introduces the concepts essential to modeling data with the SAP Infinitelnsight The shorter Chapter 4 General Introduction to Scenarios provides a summary of the application scenarios for features Infinitelnsight Modeler Regression Classification and Infinitelnsight Modeler Segmentation Clustering It also introduces the user interface and the data files used in these scenarios Chapters 5 and 6 Generating Explanatory and Predictive Models with Infinitelnsight Modeler Regression Classification feature and Generating Descriptive Models with Infinitelnsight Modeler Segmentation Clustering feature present the Infinitelnsight Modeler Regression Classification feature and the Infinitelnsight Modeler Segmentation
93. For some type of data you will need a specific license 3 Use the Browse button corresponding to the Folder field to select the folder or database containing the data In case of a protected database you will need to enter the user name and the password in the fields User and Password 4 Click the button Edit Variable Poo Content to edit the parameters of the variables stored in the variable pool 5 Click OK to validate 5 2 Creating a Classification Model Using Infinitelnsight Modeler Data modeling with Infinitelnsight Modeler Regression Classification is subdivided into four broadly defined Stages 1 Defining the Modeling Parameters 2 Generation and Validation of the Model 3 Analysis and Understanding of the Analytical Results 4 Using a Generated Model 5 2 1 Step 1 Defining the Modeling Parameters To respond to your business issue you want to Identify and understand the factors that determine whether a prospect reacts positively or negatively to your marketing campaign Thereby be able to predict the behavior of new prospects with respect to your campaign The Infinitelnsight Modeler Regression Classification feature formerly known as K2R allows you to create explanatory and predictive models The first step in the modeling process consists of defining the modeling parameters 1 Select a data source on page 67 to be used as training data set Describe the data set on page 70 selected Select
94. Infinitelnsight Modeler Regression Classification Each variable is described by the fields detailed in the following table The Field Gives information on Name the variable name which cannot be modified Storage the type of values stored in this variable Number the variable contains only computable numbers be careful a telephone number or an account number should not be considered numbers String the variable contains character strings Datetime the variable contains date and time stamps Date the variable contains dates Value the value type of the variable Continuous a numeric variable from which mean variance etc can be computed Nominal categorical variable which is the only possible value for a string Ordinal discrete numeric variable where the relative order is important Jextual textual variable containing phrases sentences or complete texts A Caution When creating a text coding model if there is not at least one textual variable you will not be able to go to the next panel Key whether this variable is the key variable or identifier for the record Othe variable is not an identifier primary identifier 2 secondary identifier Order whether this variable represents a natural order 0 the variable does not represent a natural order 1 the variable represents a natural order If the value is set at 1 the variable is used in SQL expressions in an order by condition There must
95. InsightVx y z Samples Census The following table describes those files File Name Description When is it Used Census0l csv Data file This file is used for both application scenarios used in this manual desc_censusOl cs_ Description file for the This file is used for both application scenarios used in this manual v CensusO1 csv file To obtain a detailed description of the Census01 csv file see Introduction to Sample Files see page 60 CUSTOMER SAP Infinitelnsight 7 0 8 2014 SAP AG or an SAP affiliate company All rights reserved Welcome to this Guide Documentation Full Documentation Complete documentation is included with SAP Infinitelnsight This documentation covers The operational use of SAP Infinitelnsight features The architecture and integration of the SAP Infinitelnsight API The SAP Infinitelnsight Java graphical user interface Contextual Help Each screen in the Infinitelnsight modeling assistant is accompanied by contextual help that describes the options presented to you and the concepts required for their application VI To Display the Contextual Help In the Help menu located at the upper left corner of Infinitelnsight modeling assistant click the option Help You can also directly press F1 on your keyboard to display quickly the contextual help CUSTOMER SAP Infinitelnsight 7 0 9 2014 SAP AG or an SAP affiliate company All rights reserved Welcome to this Guide 3
96. P Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Classification Decision The screen Classification Decision allows you to select how many observations you want the model to detect after application on the new data set vV To apply a Classification Decision 1 Onthe screen Applying the Model follow all the steps of the procedure To Apply a Model to a New Data Set 2 In the Generate drop down list select the option Decision 3 Click the Apply button The screen Classification Decision appears a SAP InfiniteInsight VX y 2 class_Census01 File Help Classification Decision Target class ie Threshold a of Population of Detected Target Score Threshold 23 6 Confusion Matrix Predicted 1 2941 Predicted 0 9520 True 1 2973 a Fer True 0 9488 a Total Population 124615 Cost Matrix Predicted 1 Predicted 0 True 4 True 0 4 Use the slide to set the percentage of population to detect 5 Click Next The model is applied to the new data set CUSTOMER of Population 23 6 5 of Detected Targ et 67 5 Score Threshold 0 199 Metrics Classification Rate 84 77 Sensitivity 67 54 Specificity 90 17 Precision 68 26 Fi Score 0 679 Profit 0 Random 0 Maximize Profit Gain 0 SAP Infinitelnsight 7 0 162 2014 SAP AG or an SAP affiliate company All rights re
97. Point Report Type Model Overview hai Ce E PO E O ae Number of Records 46 842 Building Date 2014 05 06 11 55 02 Learning Time 225 Engine Name Kxen SmantSeqmenter Author natacha yam Minimum Requested Number of Clusters 10 Maximum Requested Number of Clusters 10 SQL Expressions enabled Nominal Targets TargetKey 1 O0 Frequency 6 05 1 Frequency 23 95 Performance Indicators Target class Predictive Power Kl 0 740 Prediction Confidence KR 0 993 44 Cancel 41 Previous 1 Note As a general note other indicators are provided in addition to the predictive power and the prediction confidence during the generation of the model For example you could view the Learning Time required to generate the model and information on the targets CUSTOMER SAP Infinitelnsight 7 0 230 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering You have two options On the screen Training the Model you can 231 You can also verify the indicators in the Detailed Log click the Z Show Detailed Log button The following screen appears amp SAP InfiniteInsight Vx y z class_Census01 File Help Training the Model Stop ViewType Copy Print Save _Exportto ba g PowerPoint class Census01 2074 05 06 11 54 40 Prototypes optimization pass 1 4 Prototypes optimization pass 2 4 Prototypes optimization pass 3 4 Prototypes optimization pa
98. R SAP Infinitelnsight 7 0 25 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 6 2 Example In a database containing information about your customers the name and address of those customers are examples of variables 4 6 3 Types of Variables There are three types of variables Continuous variables Ordinal variables Nominal variables Continuous Variables Definition Continuous variables are variables whose values are numerical continuous and sortable Arithmetic operations may be performed on these values such as determination of their sum or their mean Example The variable salary is a numerical variable but in addition is also a continuous variable It may for instance take on the following values 1 050 1 700 or 1 750 The mean of these values may be calculated Continuous Variables and Modeling During modeling a continuous variable may be grouped into significant discrete bins CUSTOMER SAP Infinitelnsight 7 0 26 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Ordinal Variables Definition Ordinal variables are variables with discrete values that is they belong to categories and they are sortable Ordinal variables may be Numerical meaning that its values are numbers They are therefore ordered according to the natural number system O 1 2 and so on Textual meaning that its values are ch
99. R SAP Infinitelnsight 7 0 101 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Enabling the Post Processing This section allows setting some regression parameters according to three strategies This option can only be activated when the model contains at least one continuous target variable The description of these strategies and an example of performance curve for each strategy are provided in the table below Regression Strategy Description Without post processing With original target encoding With uniform target encoding CUSTOMER The first strategy consists in disabling the regression post processing during the learning model phase in order to create a regression similar to the one used in versions prior to 3 3 2 In this case a standard regression is performed No special improvement is made to the final scores Original target values are S ee eae used and raw score values are produced as outputs The second strategy which applies to regressions using a post processing consists in using the original target value during the learning model phase to compute regression coefficients The result of the regression is zi then transformed to align target segment means and score segment means in the post processing phase l Note This is the default strategy used in Infinitel nsight The last strategy which applies to regr
100. SAP InfiniteInsight VX y 2 class_Census01 File Help Selecting Contributory Variables Targets cass lt Eune View Reports Variable Maximum Contribution k KR Ki Selec r Priority Last Iteration marital status I 25 61 0532 0989 1521 O amp M a capital gain x 17 71 018 0 988 14167 O0 amp MN 6 occupation x 13 42 0 457 0 994 1 451 Tl 6 education num B 10 98 0 436 0 985 1 421 Tit 6 age WE 7 29 0 411 0 997 1 408 Til 5 capitalloss qq 6 29 0 068 0 999 1 067 amp itl 3 hours per we NE 5 79 0 345 0 991 1 336 amp tll 4 education E 5 24 0 436 0 985 1 421 Ti 2 relationship E 3 01 0 553 0 988 1 544 Tit 1 sex 2 12 0 23 0 991 1 22 itl 1 workclass ff 1 1 0 176 099 1 166 amp itl 1 native country f 0 73 0 048 0 995 1 043 amp Tit 1 race J 0 63 0 069 0 994 1 063 amp Tit 1 fniwat 0 2 0 038 0 995 1 033 amp Tit 1 Number of Selected Variables 0 44 Cancel Ej You must select at least one variable 41 Previous I Next 2 Inthe Targets list select the target variable for which you want to select the contributory variables 3 Click the button Smart Selection The window Smart Variables Selection opens ewe iteration ill K Smart Variables Selection Remark 0 variable s automatically excluded CUSTOMER SAP Infinitelnsight 7 0 181 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regressio
101. SAS v6 or 7 8 for Windows or UNIX SAS Transport to save the model in a generic SAS compatible file Depending upon which option you selected this field allows you to specify the ODBC source the memory store or the folder in which you want to save the model This field allows you to enter the name of the file or table that is to contain the model The name of the file must contain one of the following format extensions txt text file in which the data is separated by tabs or csv text file in which the data is separated by commas SAP Infinitelnsight 7 0 226 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 6 2 2 Step 2 Generating and Validating the Model Once the modeling parameters are defined you can generate the model Then you must validate its performance using the predictive power KI and the prediction confidence KR If the model is sufficiently powerful you can analyze the responses that it provides in relation to your business issue see Step 3 Analyzing and Understanding the Model Generated see page 113 and then apply it to new data sets see Step 4 Using the Model see page 154 Otherwise you can modify the modeling parameters in such a way that they are better suited to your data set and your business issue and then generate new more powerful models Generating the Model vV To Generate the Model 1 On the screen Specific Param
102. Sets The following table defines the roles of the three data sub sets obtained using cutting strategies The data set Is used to Estimation Generate different models The models generated at this stage are hypothetical Validation Select the best model among those generated using the estimation sub set which represents the best compromise between perfect quality and perfect robustness Test Verify the performance of the selected model on a new data set To understand the role of cutting strategies in the model generation process see the figure Generating a Model see page 36 4 4 3 Types of Cutting Strategies To generate your models there are two types of cutting strategies that you may use The customized cutting strategy he automatic cutting strategies Customized Cutting Strategy Definition The customized cutting strategy allows you to define your own data sub sets To use this strategy you must have prepared before opening SAP Infinitelnsight features three sub sets the estimation validation and test Sub sets CUSTOMER SAP Infinitelnsight 7 0 20 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts How to Use this Before opening the SAP Infinitelnsight cut your initial data file into three files of the size of your choice For example The first file may contain the first 1 500 observations or lines of your initial data file The second file observati
103. The variable moves to the screen section Target Variables Also select a variable in the screen section Target Variables and click the button lt to move the variables back to the screen section Explanatory variables selected CUSTOMER SAP Infinitelnsight 7 0 93 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Weight Variable Selecting a Weight Variable enables to set the Weight Quantum option available in the Advanced Model Parameters For this Scenario Do not select a weight variable To Select a Weight Variable 1 On the screen Selecting Variables in the section Explanatory variables selected left hand side select the variables you want to use as a Weight Variable fe SAP InfiniteInsight Vx y z New Regression Classification Model File Help Selecting Variables Explanatory Variables Selected 14 Target Variables 1 workclass fniwagt education education num marital status occupation relationship race sex _ Alphabetic Sort capital gain Weight Variable 0 capital loss hours per week native county Exchuded Variabli 1 EJ _ Alphabetic Sort _ Alphabetic Sort l Note On the screen Selecting Variables variables are presented in the same order as that in which they appear in the table of data To sort them alphabetically select the option Alphabetic sort presented beneath each of the variables list 2 Click th
104. To sort them alphabetically select the option Alphabetic sort presented beneath each of the variables list 2 Click the button gt located on the left of the screen section Weight Variable middle right hand side 220 CUSTOMER 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering The variable moves to the screen section Weight Variable Also select a variable in the screen section Weight Variable and click the button lt to move the variables back to the screen section Explanatory variables selected SAP Infinitelnsight 7 0 Explanatory Variables By default and with the exception of key variables all variables contained in your data set are taken into consideration for generation of the model You may exclude some of these variables The decision whether to include or exclude a variable for generation of your segmentation model depends upon domain specific considerations Your domain specific knowledge allows you to determine which variables are the most useful for description of the clusters or homogeneous groups A regression model generated using Infinitelnsight Modeler Regression Classification formerly known as K2R would also be used as a tool to determine the variables with the greatest explanatory power for a given phenomenon For this Scenario Exclude the variable Kx ndex as this is a key variable Since the initial data set does not contain a key
105. _3 and so on Outputs by Reference Category score This option allows you to generate in the output file the score corresponding to each data set line for the different categories of the target variable You can generate the scores for all the target variable categories or select specific categories It appears in the output file as rr_ lt Jarget Variable gt for the target variable key category and rr_ lt arget Variable gt _ lt Category gt for its other categories vV To Add the Score of All Target Variable Categories Check the All option VI To Add Only the Scores of Selected Categories 1 Check the Individual option 2 Inthe Selection column check the boxes corresponding to the categories for which you want to add the score in the output file Prediction Probability This option allows you to generate in the output file the probability for one or more target variable categories that is for each observation the probability of the target variable value to be the selected category It appears in the output file as proba_rr_ lt Target Variable gt for the target variable key category and as proba_rr_ lt Traget Variable gt _ lt Category gt for the other categories of the target variable V To Add the Probabilities of All Target Variable Categories Check the All option vV To Add Only the Probabilities of Selected Categories 1 Check the ndividual option 2 Inthe Selection column check the boxes corresponding to the categorie
106. a Access SAP Infinitelnsight accepts many types of data sources Flat files such as csv files files of text tables and other files of type text ODBC compatible sources such as Oracle SQL Server or IBM DB2 databases In addition the C Data Access Application Programming Interface allows you to connect proprietary format sources such as industrial sensor streams In most cases and particularly if you are using SAP Infinitelnsight features via a graphical interface you never have to concern yourself with the data access process Data access is accomplished in a semi transparent manner from the graphical user interface you need only select the data source format to be used flat files or ODBC compatible data sources and specify the location of the data file The C Data Access Application Programming Interface is helpful to developers who want to write access code for proprietary format databases CUSTOMER SAP Infinitelnsight 7 0 12 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight The Infinitelnsight Access Feature The Infinitelnsight Access formerly known as KAA feature allows reading SAS data and writing the scores obtained with an SAP Infinitelnsight model into a SAS table The following formats are currently supported SAS files version 6 under windows amp unix SAS 7 8 under windows amp unix SAS Transport Files You can access directly a SAS table with the S
107. affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Risk Mode Tab This tab allows you to select a specific learning mode for your model vV To Enable the Risk Mode 1 Select the tab Risk Mode a SAP InfiniteInsight VX Yy 2 class_Census01 File Help Advanced Model Parameters General Auto selection MiLa Gain Chart ill VvjEnable Risk Score 615 for good bad odds ratio of fo to 1 Points to double odds 15 View Score Table Risk Fitting Domain Use Score Bin Frequency as Weights cy Ox 2 Check the box Enable The tab activates and the Risk Mode settings are displayed setting Risk Mode Infinitelnsight Risk Mode allows advanced users to ask a classification model to translate its internal equation obtained with no constraints into a specified range of scores associated with good bad odds ratio When this mode is activated the different encodings that are used internally for continuous and ordinal variables are merged in a single representation allowing a simpler view of the model internal equations This is particularly useful when the usage of predictive model is subject to legal restrictions the model equations are now simple enough to be understood by legal departments and can be exposed not only in programming language as it was already the case before but even in simple words The underlying technology is also used to display so called score cards
108. affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 6 2 4 Step 4 Using the Model Once generated a clustering model may be saved for later use A clustering model may be applied to additional data sets The model thus allows you to assign observations to clusters This part presents the option Applying the model to a new data set for the Infinitelnsight Modeler Segmentation Clustering feature The other options for deployment of the clustering models are similar to those proposed for models generated using the Infinitelnsight Modeler Regression Classification feature For more information about these options see Saving the Model Opening the Model Applying the Model to a New Data Set The currently open model may be applied to additional data sets The model allows you to determine to which cluster the observations described in these data sets belong Constraints of Model Use In order to apply a model to a data set the format of the application data set must be identical to that of the training data set used to generate the model The same target variable in particular must be included in both data sets even if values for the target variable are not contained in the application data set CUSTOMER SAP Infinitelnsight 7 0 263 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Using the Option Direct Apply in the
109. ain a variable that serves as a key variable Two cases should be considered If the initial data set does not contain a key variable a variable index Kx ndex is automatically generated by SAP Infinitelnsight features This will correspond to the row number of the processed data If the file contains one or more key variables they are not recognized automatically You must specify them manually in the data description See the procedure To Specify that a Variable is a Key see page 75 To Specify that a Variable is a Key 1 Inthe Key column click the box corresponding to the row of the key variable 2 Type in the value 1 to define this as a key variable a A Fi SAP I InfiniteInsight Vx y Z New Regression Classification Model Miel Data Description Main Edition Structures 3 w Open Description a Save in Variable Pool Y ee el Save Description Remove from Variable Pool Description 6 Description Desc_Census01 csv a a a lage ee a E O TEE EE ORE E TE 3fnwgt umber continuous 00 l oOo 4education sting nominal 0 0 Z o S y O O i Seducation n number ordinal 0 0 2 O S S d o o 6 marital status sting nominal 0 Toccupation sting nominal 0 0 Z P OO o o O ai amp relationship sting nominal 0D d o o race sting nominal 0 Q O o S d O oo 10sex sting nomina 0 o S S S S O yoo 11 capital gain number continuous 0 99999 o o oo 12 capitaoss number continuou
110. ains while none of the customers contained in cluster 6 fail to realize some annual capital gain Checking the Fix Variable box would allow you to compare the profiles of the variable capital gain for all the segments SQL Expressions The Cross Statistics screen also allows you to visualize the SQL Expression used to define each cluster l Note SQL Expressions are only available if you have selected the Calculate SQL Expressions option in the Modeling Parameters Advanced Screen before generating your model CUSTOMER SAP Infinitelnsight 7 0 257 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering vV To Display SQL Expression for a Cluster 1 Select the cluster in the summary table The plot for the selected cluster is displayed 2 Click View Type and select the gF SQL button The SQL expression replaces the cross statistics plot in the lower part of the screen a SAP Infinite Insight Vx y z cdass_Census01 File Help aly Cluster Profiles k Gade E View Type Bar Sort Reset Cluster Copy Print Save Export to eee Orientation Categories Names lt Excel Cluster Name Frequencies of 1 1 f 4 22 2 RE 25 24 3 715 4 F 2 44 5 E 11 32 T Variable ranges for Cluster 6 i capital gain in KxMissing or 4366 41310 44 Cancel 41 Previous Ik OK 3 Click the small magnifier icon to explore the SQL express
111. al user defined comment Character string t that can be used to identify the model 5 Select a model from the list 6 Click the Open button The screen Using the Model appears 4 SAP InfiniteInsight VX y 2 class_Census01 File Help Ye Using the Model Display uh Display Run Model Overview Model Graphs Save Export Contributions by Variables Category Significance Statistical Reports Scorecard Confusion Matrix 44 Cancel 4 Previous CUSTOMER SAP Infinitelnsight 7 0 196 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 6 Infinitelnsight Modeler Segmentation Clustering IN THIS CHAPTER Application Scenario Customize your Communications using Data Modeling cceeeeeeeeeeeeeeeeeeeeeeeeeeeeeaeees 197 Creating a Clustering Model Using Infinitelnsight MOdeler ccccccccsseeeeeeeeeeeeeeeeeeeeeeeeeeeaaneeeeeesseeeseneeeeeees 207 6 1 Application Scenario Customize your Communications using Data Modeling In this scenario you are the Marketing Director of a large retail bank The bank wants to offer a new financial product to its customers Your project consists of launching a direct marketing campaign aimed at promoting this product In order to customize the marketing messages from the bank and improve communication with the various customers and prospects for this new product the senior management of the ba
112. and Feel Display 3D Chart Disable Double Buffering Optimize for Remote Display Remember Size and Position when Leaving Report Number of Variables of Interest Active Style Sheet Customize Style Sheets Customizing Style Sheets SAP Infinitelnsight offers the possibility to customize the generated reports The default style sheet called SAP Infinitelnsight Report Style Sheet default cannot be modified You have to create your own style sheets to modify the settings l Note To create load or save a style sheet you have to indicate a data source in the panel Edit Options before opening the window SAP Infinitelnsight Report Style Sheet Editor CUSTOMER SAP Infinitelnsight 7 0 204 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering VI To Create a New Style Sheet In the field Folder click the button J Browse 2 Selecta folder This folder is your style sheets repository 3 Click the button oi Add A new style sheet has been created 4 Click the button id The panel Report Style Sheet Editor opens 5 Inthe field Style Sheet Name enter a name for the new style sheet The extension krs is automatically added l Note You can duplicate a style sheet by changing the name of your style sheet The previous one is not deleted CUSTOMER SAP Infinitelnsight 7 0 205 2014 SAP AG or an SAP affiliate company All rights reserve
113. any All rights reserved Glossary KL Kullback Leibler The Kullback Leibler divergence is used to measure the difference between the cluster profile and the population profile of the variables KPI Key Performance Indicator KPIs or key performance indicators help organizations achieve organizational goals through the definition and measurement of progress The key indicators are agreed upon by an organization and are indicators which can be measured that will reflect success factors The KPIs selected must reflect the organization s goals they must be key to its success and they must be measurable K S test K S is the Kolmogorov Smirnov statistic applied here as a measure of deviation from uniform response rates across categories of a variable Kolmogorov Smirnov is a non parametric exact goodness of fit statistic based on the maximum deviation between the cumulative and empirical distribution functions L Lift The Lift of a rule is a measure that indicates the chances of finding the consequent by using the antecedent compared with the chances of randomly finding the consequent A value greater than 1 indicates that using the antecedent increases your chances to find the consequent Lift profit Lift profit allows examination of the difference between a perfect model and a random model and between the model generated by SAP Infinitelnsight and a random model It represents the ratio between a model and the random model t
114. any All rights reserved Infinitelnsight Modeler Regression Classification aj This option allows you to select which columns to display for current report Series Usage Options This option allows you to copy the data from the current view of the displayed report The data can then be pasted in a text editor a soreadsheet a word processing software This option allows you to print the current view of the selected report depending on the chosen display mode HTML 3 table graph D P This option allows you to save under different formats text html pdf rtf the data from the current view of the selected report This option allows you to save under different formats text html pdf rtf the data from all the views of the selected report This option allows you to export to Excel This option allows you to save all reports This option allows you to save the customized style sheet CUSTOMER SAP Infinitelnsight 7 0 141 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Scorecard This screen provides you with the coefficients associated to each category for all variables in a regression model V To Obtain a Score Add all the coefficients corresponding to the selected value of each variable Category L Note 1 968e 2 7 913e 2 2 331e 4 age 7 517e 2 3 614e 4 age 8 527e 2 1 312 3 age 1 043e 1 22 4e0 3 agel 1 892
115. ap con corporate en legal copyright i ndex epx trademark for additional trademark information and notices statements accompanying such products and services P d ta O Le A
116. aphical interfaces CUSTOMER SAP Infinitelnsight 7 0 11 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight The KxShell Command Interpreter The KxShell command interpreter allows you to use SAP Infinitelnsight by typing commands or executing scripts containing several commands The command interpreter is an example of development based on the C API Like any other API it may be used to integrate SAP Infinitelnsight with other applications or program packages Control API The Control API Application Programming Interface is aimed primarily at developers or users with programming experience This Application Programming Interface is used to access the complete range of functionalities and the most fine grained parameterization of SAP Infinitelnsight features In addition it allows customized integration of SAP Infinitelnsight features with other applications or program packages Three APIs are provided with SAP Infinitelnsight ACOM DCOM API usable over Microsoft platforms A CORBA API usable over all client server platforms A C API usable over all standalone platforms 3 2 2 Operations The operation of SAP Infinitelnsight may be subdivided into four phases Phase 1 Data access on page 12 Phase 2 Data manipulation and preparation on page 13 Phase 3 Data modeling on page 14 Phase 4 Model presentation and deployment on page 14 Phase 1 Dat
117. ar is a deviation measure of the values around the predicted score Possible values are 1 if the observation is an outlier with respect to the current target else O Contributions contrib_ lt variable gt _rr_ lt target add the variables contributions for the current variable to the variable gt output file You can add the contributions of all variables or select only the contributions of specific variables see For example if marital status is an explanatory procedure below variable for the target variable class the column contrib_marital status_rr_class will be generated in the output file V To Add All Variables Contributions Check the All option vV To Add Specific Variable Contributions 1 Check the ndividual option 2 Click the gt gt button to display the variable selection table 3 Inthe Available list select the variables you want to add use the Ctrl key to select more than one variable 4 Click the gt button to add the selected variables to the Selected list Predicted Value This option is checked by default It allows you to generate in the output file the value predicted by the model for the target variable It appears in the output file as rr_ lt target variable gt Outlier Indicator This option allows you to show in the output file which observations are outliers An observation Is considered an outlier if the difference between its predicted value and its real value exceeds the value of the
118. aracter strings They are therefore ordered according to alphabetic conventions Example The variable school grade is an ordinal variable Its values actually belong to definite categories and can be sorted This variable can be numerical if its values range between O and 20 textual if its values are A B C D E et F A Caution A variable assessment which values are good average and bad cannot be directly treated as an ordinal variable by SAP Infinitelnsight features The values would be sorted in alphabetical order C average bad good and not according to their meaning When a nominal variable order is important the variable must be encoded in letters or numbers before it can be used by SAP Infinitelnsight Nominal Variables Definition Nominal variables are variables whose values are discrete that is belong to categories and are not sortable Nominal variables may be Numerical meaning that its values are numbers Textual meaning that its values are character strings A Caution Binary variables are considered nominal variables CUSTOMER SAP Infinitelnsight 7 0 2 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Example The variable zip code is a nominal variable The set of values that this variable may assume 10111 20500 90210 for example are clearly distinct non ranked categories although they happen to be represent
119. aracteristics as the training data set CUSTOMER SAP Infinitelnsight 7 0 115 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Options M M 116 To Copy the Model Summary Click the A Copy button The application copies the HTML code of the screen You can paste into a word processing or spreadsheet program a text editor To Save the Model Summary Click the Save button situated under the title The file is saved in HTML format To Print the Model Summary Click the J Print button situated under the title A dialog box appears allowing you to select the printer to use Select the printer to use and set other print properties if need be Click OK The report is printed To Export to PowerPoint PPT Export to Click the PowerPoint Export to PowerPoint button CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Model Graphs Displaying the Model Graph vV To Display the Model Graph 1 Onthe screen Using the Model click the Model Graphs option The model graphs appears When the target is nominal the following curve is displayed Profit Type Detected Vv Models rr_class v Performance 0 9 4 0 8 0 7 4 0 6 0 5 4 Detected Profit 0 4 0 3 4 0 1 5 0 0 oh ph or ar no ap ap ape or oP er
120. arget Ratio that is the percentage of the node population for which the target is positive Negative Target Count that is the number of records for which the target Is negative Negative Target Ratio that is the percentage of the node population for which the target is negative Variance that is the variance for the current node Weighted Population that is the number of records when using a weight variable CUSTOMER SAP Infinitelnsight 7 0 150 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Profit Curve The profit curve for the current decision tree is displayed in the tab Profit Curve located in the lower part of the panel This profit curve changes with every modifications made on the decision tree Node Details Profit Curve oe on a nso ta The profit curve corresponding to the node containing the whole population is equal to the random curve CUSTOMER SAP Infinitelnsight 7 0 151 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Profit Curve E Random E Wizard E Decision Tree E KZ2R Model oe Ko ie To te ho he th he To Te fo Te To To ho Te Fo th When you expand the node with the highest percentage of positive target the profit curve improves over the first percentiles which means that the model will detect the population with the high
121. ariable web address This variable contains the Web site address of the corporate customers contained in the database Some companies have a Web site others do not In addition each Web site address is unique In this case SAP Infinitelnsight automatically transforms the web address variable into a binary variable with two possible values KxOther the firm has a Web site and KxMissing the firm does not have a Web site Statistical Reports Options A tool bar is provided allowing you to modify how the current report is displayed to copy the report to print it to save it or to export it to Excel Statistical Reports The Statistical Reports provide you with a set of tables that allows you a more detailed debriefing of your model These reports are grouped in different levels of debriefing the Descriptive Statistics which provides the statistics on the variables their categories and the data sets as well as the variables cross statistics with the target l Note If your data set contains date or datetime variables automatically generated variables will appear in the statistical reports For more information refer to section Date and Datetime Variables Automatically Generated Variables on page 30 In the section Cross Statistics with the Target s the number of displayed categories corresponds to the number of categories as defined in the user structure the band count if no user structure has been defined For
122. ariable is associated with too many missing values the missing values are grouped in the KxMissing category that is also created automatically To understand the value of the categories KxOther and KxMissing consider the following example The database of corporate customers of a business contains the variable web address This variable contains the Web site address of the corporate customers contained in the database Some companies have a Web site others do not In addition each Web site address is unique In this case SAP Infinitelnsight automatically transforms the web address variable into a binary variable with two possible values KxOther the firm has a Web site and KxMissing the firm does not have a Web site CUSTOMER SAP Infinitelnsight 7 0 246 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Clusters Summary The following types of charts can be displayed Bubble Charts Bubble charts display the clusters by representing the relationship between three variables Bar Charts Cluster Plots Bar charts displays the three cluster plots that allow you to examine The proportion of observations of the data set contained in each cluster Frequencies plot The proportion of each cluster relative to the target variable Target Means and Relative Target Means plots Displaying the Bubble Charts vV To Display the Bubble Charts 1 On the screen Usi
123. ariables Automatically Generated Variables on page 30 the Model Performance in which you will find the model performance indicators the variables contributions and the score detailed statistics the Control for Deviations which allows you to check the deviations for each variable and each variable category between the validation and test data sets the Expert Debriefing in which you will find more specialized performance indicators as well as the variables encoding the excluded variables during model generation and the reason for exclusion and so on Statistical Reports Options A tool bar is provided allowing you to modify how the current report is displayed to copy the report to print it to save it or to export it to Excel CUSTOMER SAP Infinitelnsight 7 0 261 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Display Options E view Hi 3 af Sort ally A 5 i W 32 v a This option allows you to select which columns to display for current report Series This option allows you to display the current report view in the graphical table that can be sorted by column This option allows you to display the current report view as a HTML table Some reports can be displayed as a bar chart This bar chart can be sorted by ascending or descending values or by ascending or descending alphabetical order You can also select which
124. ariables list For more information refer to section Date and Datetime Variables Automatically Generated Variables on page 30 1 Note You can display the relative significance of the categories of a variable directly from the plot Contributions by Variables On the plot Contributions by Variables double click the bar of the variable which interests you In case no user structure has been defined for a continuous variable the plot category significance displays the categories created automatically using the band count parameter The number of categories displayed corresponds to the value of the band count parameter For more information about configuring this parameter please refer to the section Band Count for Continuous Variables CUSTOMER SAP Infinitelnsight 7 0 129 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Plot Options VI To Switch Between Validation Data Set and All Data Sets Plots 1 Click Data Sets and select the 29 All Data Sets button to display all data sets The plot displaying all data sets appears a SAP Infinite Insight Vx y z cass_Census01 File Help aly Category Significance A w 9 Asb 4 a Data Sets View Type Bar Copy Print Save EXPortto pin View ee fkt r Orientation Excel Variables age kal Variable age Influence on Target 0 075 0 050 0 025 0 000 0 025 0 050 0 075 0 100 0 125 0 150 0 17
125. ars vV To Display the Detailed Log Click View Type and select the Log button The following screen appears a SAP InfiniteInsight VX y 2 class_Census01 File Help Training the Model a 8 eu Stop View Type Copy Print Save Export to eee nce PowerPoint class Census01 2074 05 06 09 74 48 Computing statistics Indicator on validation Quality KI 0 746901 Robustness HR 0 9927623 Standard errors on estimation L1 0 300967 L2 0 560722 Lint 0 990842 Standard errors on validation L1 0 302877 L2 0 362271 Lint 0 93 7751 Index of current iteration 6 Reference EI 0 808648 Reference ER 0 995696 Number of kept variables 3 KI 0 746901 KR 0 992623 For the current iteration 3 variables kept KI 0 746901 KR 0 992623 Computing statistics Computing statistics Indicator on validation Quality KI 0 808646 Robustness KR 0 995696 Standard errors on estimation L1 0 266717 L2 0 3534585 Lint 1 24211 Standard errors on validation L1 0 2668897 L2 0 335906 Lint 1 06301 Learning time 18 seconds Classification regression learning phase finished Total elapsed time 19 seconds End of the training process Cs Oy CUSTOMER SAP Infinitelnsight 7 0 228 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering vl To Stop the Learning Process cr a Click the wE Stop Current Task button 2 Click the Previous button The
126. as a risk score of about 15 According to the parameter PDO set in this example to 15 it is easy to conclude that the segment 37 43 is two times more risky or that the odds of the segment 37 43 are two times inferior to the segment 24 27 i KXEN InfiniteInsight class_Census01 allt Score Card Category VF 24 28 21 34 ar 44 53 ey 90 Defaut CUSTOMER SAP Infinitelnsight 7 0 143 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Confusion Matrix The panel Confusion Matrix allows you to visualize the target values predicted by the model compared with the real values and to set the score above which the observations will be considered as positive that is the observations for which the target value is the one wanted This panel also allows you to simulate your profit depending on the selected threshold score or to automatically adapt the threshold to obtain a maximum profit 4 SAP InfiniteInsight VX y 2 cass_Census01 File Help alli Confusion Matrix Target class ili Threshold 2 of Population of Detected Target Score Threshold of Population 23 6 5 of Detected Targ et 67 5 93 6 Score Threshold 0 199 Confusion Matrix Metrics P edicted 1 2941 P edicted 0 9520 tte S aaa at ened ital et Sensitivity 67 54 True 1 2973 fi Fe end Specificity 90 17 a A ae Prec
127. assification 5 1 7 Introduction to Sample Files This guide is accompanied by the following sample data files A data file CensusOl csv The corresponding description file desc_census csv These files allow you to evaluate SAP Infinitelnsight features and take your first steps in using it CensusOl csv is the sample data file that you will use to follow the scenarios of Infinitelnsight Modeler Regression Classification and Infinitelnsight Modeler Segmentation Clustering This file is an excerpt from the American Census Bureau database completed in 1994 by Barry Becker l Note For more information about the American Census Bureau see http www census gov CUSTOMER SAP Infinitelnsight 7 0 60 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification This file presents the data on 48 842 individual Americans of at least 17 years of age Each individual is characterized by 15 data items These data or variables are described in the following table Variable age workclass fniwgt education education nu m marital status occupation relationship race SEX capital gain capital loss native country class 1 Note Description Age of individuals Employer category of individuals Weight variable allowing each individual to represent a certain percentage of the population Level of study represented by a schooling level o
128. at an alphanumeric value such as occupation is missing The value 99999 means that a numerical value such as age is missing Unfortunately you have neither the time nor the necessary resources to Performa survey to fill in the missing information Re format the database Technical Environment The database available to you is stored in an RDBMS relational database management system residing ona UNIX server maintained by the Information Technology department of the bank The technical constraints of this information environment are determining factors in selecting potential data analysis tools 5 1 4 Your Approach By virtue of the critical stakes involved in this campaign because of your limited budget and your inability to predict customers enthusiasm for the new product you have chosen to minimize your risks by dividing the project into two steps 1 Test the marketing campaign on a sample of 50 000 individuals extracted from the prospects database of 1 000 000 people 2 Global launch of the marketing campaign using the entire contents of the prospects database The Test Phase of Your Marketing Campaign The test phase of your marketing campaign allowed you to collect a sample of 50 000 individuals whose behavior with respect to this new product is known 25 of the prospects showed themselves to be clearly interested They chose to accept an invitation for a meeting with one of your sales channel agents
129. at is to contain the model The name of the file must contain one of the following format extensions txt text file in which the data is separated by tabs or csv text file in which the data is separated by commas CUSTOMER SAP Infinitelnsight 7 0 192 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Files Created when Saving a Model When you save a model SAP Infinitelnsight creates a series of files in the specified store The following table lists the files or tables created when saving a model and in which case Filename Description Used by KxAdmin lists all the models contained in the folder database with additional all models created with information date version name of the model comments Infinitel nsight lt Model_name gt file named after the model and containing all the model data except all models created with graphs information Graphs are stored in additional tables see below _ Infinitelnsight KxInfos indicates which additional tables are needed by the model all models created with Infinitel nsight KxOlapCube Stores the OLAP Cube used by the decision tree when the option Infinitel nsight Modeler Infinitel nsight Modeler Regression Classification as Decision is Regression Classification activated models with decision tree KxLinks contains the links from the graphs of the model Infinitel nsight Social models only KxNodes lists all the nodes f
130. ated in the panel Summary of Modeling Parameters a message is displayed at the end of the learning process confirming that the model has been saved i SAP InfiniteInsight Messages Error List mx Copy Clear Information New 5 6 14 9 25 07 AM The model class _Census01 has been saved Close 3 Click Close 4 Once the model has been generated click Next to go to panel Using the Model CUSTOMER SAP Infinitelnsight 7 0 110 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Following the Progress of the Generation Process There is two ways for you to follow the progress of the generation process M vV To Stop the Learning Process 111 The Progress Bar displays the progression for each step of the process It is the screen displayed by default The Detailed Log displays the details of each step of the process To display the Progression Bar Click View Type and select the To Display the Detailed Log Click View Type and select the E Log button The following screen appears 4 SAP InfiniteInsight Vx y 7 class_Census01 File Help Training the Model EAB Stop View Type Copy Print Save EXPOTt to A OEN i PowerPoint class Census01 2014 05 06 09 24 48 Computing statistics Indicator on validation Quality KI 0 746901 Robustness KR 0 992625 Standard errors on estimation L1 0 300967 S
131. ation data set and that of the validation data set divided by the area found between the curve of the perfect model and that of the random model Contributions by Variables Definition The Contributions by Variables plot allows you to examine the relative significance of each of the variables within the model On this plot each bar represents the contribution of an explanatory variable with respect to the target variable The following four types of plots allow you to visualize contributions by variables Variable Contributions that is relative importance of each variable in the built model Variable Weights that is weights in the final polynomial of the normalized variables Smart Variable Contributions that is the variables internal contributions Maximum Smart Variable Contributions that is the maximum smart variable contributions including only the maximum of similar variables For example only binned encoding of the continuous variable age will be displayed This is the chart displayed by default CUSTOMER SAP Infinitelnsight 7 0 125 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Displaying Contributions by Variables vV To Display the Plot of Contributions by Variables 1 On the screen Using the Model click the option Contributions by Variables The plot Contributions by Variables appears The default plot type is Maximum Smart Varia
132. ation of the data set The profit that may be achieved using a random model that does not allow bottom one to know even a Single value of the target variable for each observation of the data set CUSTOMER 123 For instance by selecting 25 of the observations from your entire data set with the help of a perfect model 100 of observations belonging to the target category of the target variable are selected Thus maximum profit is achieved l Note These 25 correspond to the proportion of prospects who responded in a positive manner to your marketing campaign during your test phase For these prospects the value of the target variable or profit is equal to 1 25 of the observations from your initial data set with the help of the model generated 66 9 of the observations belonging to the target category of the target variable are selected 25 of the initial data set using a random model 25 belonging to the target category of the target variable are selected SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification For a Model with a Continuous larget The following graph represents the model curve plot produced using a continuous target Debriefing Type Predicted a Vv Models rr_age v Actual a 22 5 25 0 27 5 30 0 32 5 35 0 37 5 40 0 42 5 46 0 47 5 50 0 52 5 55 0 57 5 60 0 62 5 Predicted Wizard Val
133. bably bring a biased information and should not be used for the modeling A special attention should be taken towards those variables A more detailed report lists which variables exactly are suspicious and at which extent see Statistical Reports gt Expert Debriefing gt Suspicious Variables Targets For each nominal variable lt Name gt Name of the target variable Target key Wanted target value Target categories Frequency Percentage of all the target value in the Estimation data set when dealing with a nominal target For each continuous target variable lt TargetName gt Name of the target variable Min Minimum value found for the target variable in the Estimation data set Max Maximum value found for the target variable in the Estimation data set Mean Mean of the target variable values on the Estimation data set Standard deviation Mean of the distance between the target values and the Mean Performance Indicators For each target rr_ lt TargetName gt Target name kc_ lt TargetName gt Note that rr_ indicates a regression classification and KC_ indicates a segmentation clustering Predictive Power KI Quality indicator that corresponds to the proportion of information contained in the target variable that the explanatory variables are able to explain Prediction Confidence Robustness indicator that signifies the capacity of the model to achieve the same performance KR when it is applied to a new data set exhibiting the same ch
134. be compared to the mean of the whole population contribution to determine which variables are the most differential Indicates whether you want to generate the reason codes when the customer variable contribution is above or below the threshold Warning Using Below with the Minimum threshold or Above with the Maximum threshold will generate an error 1 lt Target Name gt 2 Select Reason Codes 3 Click the button located on the right of the displayed table 4 Click in the cell corresponding to the parameter you want to set The following table sums up the available parameters Parameter Values Description Number of Reason Integer Codes Default 3 Threshold Mean default Maximum Minimum Criterion Below default Above 5 Ifyou want to generate several types of reason codes repeat steps 3 and 4 for each type Output The output table contains two columns for each reason code requested 168 reason_name_ lt criterion gt _ lt threshold gt _ lt rank gt _rr_ lt target name gt contains the name of the variable selected as a reason code For example the output column named reason_name_Below_Mean_1_rr_class contains the name of the variable being the most important 1 reason code with respect to the target variable class Among the variables whose contribution is below Below the mean Mean of the population contribution the selected variable will be the one having the highest deviation with it reason_value_ lt cr
135. ble Contributions a SAP InfiniteInsight VX y 2 cass_Census01 File Help ali Contributions by Variables Bsa A Bar Copy Print Save EXPOrtto pin view Orientation Excel Chart Type Maximum Smart Variable Contributions w Maximum Smart Variable Contributions 0 000 0 025 0 050 0 075 0 100 0 125 0 150 0 175 0 200 0 225 0 250 marital status capital gain occupation education num age capital loss hours perweek Variables education relationship Sex womwcelass native country face fnilwwgt If your data set contains date or datetime variables automatically generated variables can appear in this panel For more information refer to section Date and Datetime Variables Automatically Generated Variables on page 30 2 Youcan drill down on a variable that is display the plot of details of this variable where the categories of the variable can be seen To zoom in on a variable double click the corresponding bar Go to section Significance of Categories CUSTOMER SAP Infinitelnsight 7 0 126 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Understanding Contributions by Variables Only the plot Maximum Smart Contributions by Variables the default selection is presented in this guide The Contributions by Variables option allows the user to examine the relative significance of each of the explana
136. ble for report items of the view type Graphical Switch Bar Orientation this option allows having another bar orientation as the default one for a specific report item Sort by Sort Order you can select a column to sort by and choose between an ascending or a descending order Visibility you can hide columns of a report item or even menu items Note that at least one column of a report item must remain visible vV To Apply the New Style Sheet to the Generated Reports In the panel Report select the new style sheet 2 ClickOK A window opens indicating that you have to restart the modeling assistant to take the edited options into account 3 Click OK When training a model all the generated reports the learn excel statistical reports are now customized CUSTOMER SAP Infinitelnsight 7 0 206 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 6 2 Creating a Clustering Model Using Infinitelnsight Modeler Data modeling with Infinitelnsight Modeler Segmentation Clustering is subdivided into four broadly defined stages 1 Defining the Modeling Parameters 2 Generation and Validation of the Model 3 Analysis and Understanding of the Analytical Results 4 Using a Generated Model 6 2 1Step 1 Defining the Modeling Parameters To respond to your business issue you want to Break down the sample of 50 000 prospects who responded to the test phase of your marketin
137. ble gt _2 and so on until the furthest centroid You can add the distances from all centroids or only the shortest V To Add All the Distances Check the All option vV To Add Only the Shortest Distances 1 Check the Top option 2 Inthe text field enter the number of distances you want to add for example the two three or four Shortest l Note When the SQL mode is activated the notion of nearest cluster does not exist If a case belongs to a cluster distance is set to O If a case does not belong to a cluster distance is set to 1 CUSTOMER SAP Infinitelnsight 7 0 266 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Probabilities This option allows you to add to the output file the probabilities that the observation belongs to each cluster The probability for the observation to belong to the closest cluster is displayed in the column kc_best_proba_ lt largetVariable gt this probability is usually the highest The probability for the observation to belong to the second closest cluster is displayed in the column kc_best_proba_ lt TargetVariable gt _2 and so on until the furthest cluster You can add all the probabilities or only the ones corresponding to the closest clusters vV To Add All Probabilities Check the All option vV To Add Only the Probabilities for the Closest Clusters 1 Check the Top option 2 Inthe text field enter the number of probabi
138. cally select the option Alphabetic sort presented beneath each of the variables list 2 Click the button gt located on the left of the screen section Target Variables upper right hand side The variable moves to the screen section Target Variables Also select a variable in the screen section Target Variables and click the button lt to move the variables back to the screen section Explanatory variables selected CUSTOMER SAP Infinitelnsight 7 0 219 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Weight Variable 1 File Help For this Scenario Do not select a weight variable To Select a Weight Variable On the screen Selecting Variables in the section Explanatory variables selected left hand side select the variables you want to use as a Weight Variable Selecting Variables Explanatory Variables Selected 14 age workclass fnlwgt education education num marital status occupation relationship race SEX capital gain capitalloss hours perweek native county 4 Cancel Note g H _ Alphabetic Sort A SAP InfiniteInsight Vx y z New Regression Classification Model gt gt Target Variables 1 class _ Alphabetic Sort Weight Variable 0 Excluded Variables 1 H M Alphabetic Sort On the screen Selecting Variables variables are presented in the same order as that in which they appear in the table of data
139. ccceeeensssseeeeeeeeeeees 197 6 1 1 PresentaUON ociscene ih ie e aaea a T T ei Ooh aha ahead ae EE 197 6 1 2 YOU ODIECUV O denenen aa a a a A A N Re eee 198 6 1 3 YOU APD OAC erei a Wud E E E realest veteee tesa niet naa 198 6 1 4 BOO UI NS NME 51 UC ea N 198 6 1 5 Yo OO oaee E A a A A de 199 6 1 6 Mtrod ciontosample Files serrr iaa na a EA E int i Ota atk le eet eae 201 6 1 7 Himtelnsient modelng assistant eirean a a a 203 Creating a Clustering Model Using Infinitelnsight Modeler o oo cccccceesssscseeeeeeeeeeeeeeeeeeceeeessnttsssaeeeeeeees 207 6 2 1 Step 1 Defining the Modeling ParameterS sssssssssesnnirrrrrrrssssrresesrrrrrrrrrrrrrrrrssrertrrrrrrrrrrrrrrrrrre gt 207 6 2 2 Step 2 Generating and Validating the Model citi oe oi tes ie cecedede ahi ce oth toad 227 6 2 3 Step 3 Analyzing and Understanding the Model Generated 0 0 0 0 cccceeesssssssssssesssssseeeeeeeeeeeeeeeens 232 6 2 4 SEP 4 AU Shine EMOGA A NANNA ANNAA 263 GIOSS ANY aoaaa aA a i 271 CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved How to Use this Document 1 How to Use this Document IN THIS CHAPTER Organization of this DOCUMENL cccccccccssessseeecceeeeeaeeeseceeeeeeeaeeeseeeeeessaeeeeeeeeeeeeseeseeeeeeesseeasseeeeesssaaaseeeeeesssaaaaeees 4 Which Sections should YOU Read cccccccesessseecceeeceeeeseeeeeceeeeeesseceeeeeeeeeeeeceeeeeesaeeeseceeeesesaeesseceeeesssaeaeeeeeessssagasees 5
140. ced Options The panel Specific Parameters of the Model provides you with several options a SAP InfiniteInsight VX y 2 cdass_Census01 File Help Specific Parameters of the Model Calculate Cross Statistics Target Key Settings Target Target Key class Distance System Determined Encoding Strategy Target Mean 44 Cancel 4H Previous oK CUSTOMER SAP Infinitelnsight 7 0 223 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering You may calculate the cross statistics for the model to be generated define the target key value choose the distance computing option or choose the encoding strategy Calculating the Cross Statistics This option allows you to visualize the profile of each explanatory variable for each cluster with respect to their profile for the entire data set V To Enable the Cross Statistics Calculation Check the Calculate Cross Statistics box Defining the Target Key Value The Set Target Keys value option lists the target variables selected in the Selecting Variables screen and allows you to choose their key value VI To Define the Target Key Value In the Target Key field enter the key value of the target variable Choosing the Distance Computing Method The Distance list allows you to specify the distance used to compare K2C encoded input data vV To Choose the Distance Computing Method In the Distance drop
141. cenario Customize your Communications using Data Modeling 3 Going directly to the relevant section Infinitel nsight Modeler Regression Classification Infinitel nsight Modeler Segmentation Clustering You can Follow the application scenarios for a review of the features that interest you Application Scenario Enhance Efficiency and Master your Budget using Modeling Application Scenario Customize your Communications using Data Modeling Use this document as a reference text consulting it as required In this case the detailed table of contents and the index will be valuable tools helping you find the information that you seek SAP Infinitelnsight 7 0 5 2014 SAP AG or an SAP affiliate company All rights reserved How to Use this Document 1 3 Conventions Used in this Document To facilitate reading certain publishing conventions are applied throughout this guide These are presented in the following table The following information items Are presented using For example Graphical interface features and file names Arial bold Click Next The titles of particularly useful sections Garamond italicized bold See Operations The titles of procedures To Select the Target Variable The titles of sections specific to the scenario S presented in this guide For this Scenario CUSTOMER SAP Infinitelnsight 7 0 6 2014 SAP AG or an SAP affiliate company All rights reserved How to Use th
142. composed by one or more transactions simulation Application of a model to only one record smart variable contributions The variable contribution in a model while taking into account the variable correlation CUSTOMER SAP Infinitelnsight 7 0 288 2014 SAP AG or an SAP affiliate company All rights reserved Glossary social network analysis Social network analysis is used to approach problems such as community identification diffusion in graphs product adoption epidemiology graph evolution or influence of an individual within a community leader vs follower standard deviation The standard deviation is a measure of the dispersion of a collection of numbers Standardized profit Standardized profit allows examination of the contribution of the model generated by SAP Infinitelnsight features relative to a model of random type that is in comparison with a model that would only allow to select observations at random from your database This profit is used for the plots of variable details which present the significance of each of the categories of a given variable with respect to the target variable Statistical report The Statistical Reports provide you with a set of tables that allows you a more detailed debriefing of your model storage To describe the data SAP Infinitelnsight uses five types of storage formats date datetime number integer string CUSTOMER SAP Infinitelnsight 7 0
143. cted from the entire initial data set X axis On the X axis the observations are sorted in terms of decreasing score that is the decreasing probability that they belong to the target category of the target variable In the application scenario the model curves represent the ratio of prospects likely to respond in a positive manner to your marketing campaign relative to the entire set of prospects contained in your database Detected profit is the default setting for type of profit Using this type of profit The value O is assigned to observations that do not belong to the target category of the target variable The value 1 frequency of the target variable in the data set is assigned to observations that do belong to the target category of the target variable The following table describes the three curves represented on the plot created using the default parameters The curve Wizard green curve at the top Validation blue curve in the middle Random red curve at the Represents The profit that may be achieved using the hypothetical perfect model that allows one to Know with absolute confidence the value of the target variable for each observation of the data set The profit that may be achieved using the model generated by I nfinitel nsight Modeler Regression Classification that allows one to perform the best possible prediction of the value of the target variable for each observ
144. ctively orient the training process To declare a variable a weight variable results in creating a number of copies of each of the data set observations proportional to the value they possess for that variable CUSTOMER SAP Infinitelnsight 7 0 293 2014 SAP AG or an SAP affiliate company All rights reserved Glossary www sap com contactsap 2014 SAP AG or an SAP affiliate company All rights reserved No part of this publication may be reproduced or transmitted in any form or for any purpose without the express permission of SAP AG The information contained herein may be changed without prior notice Some software products marketed by SAP AG and its distributors contain proprietary software components of other software vendors National product specifications may vary These materials are provided by SAP AG and its affiliated companies SAP Group for informational purposes only without representation or warranty of any kind and SAP Group shall not be liable for errors or omissions with respect to the materials The only warranties for SAP Group products and services are _ those that are set forth in the express warranty if any Nothing herein should be construed as constituting an additional warranty SAP and other SAP products and services mentioned _ herein as well as their respective logos are trademarks or registered trademarks of SAP AG in Germany and other countries Please see www s
145. d Infinitelnsight Modeler Segmentation Clustering VI To Delete a Style Sheet 1 Select one of the displayed style sheets 2 Clickthe button Remove Note that the style sheet is not only deleted from the list but also from the data source V To Edit the General Settings Settings Options Note Reports Background Color choose a color Only the PDF and HTML formats can display a make transparent background color Edit Configuration font size Check the option Dynamically render option font style changes or click Apply when editing the settings font color so that you can visualize the result text background color table configuration The selected settings will be applied to both the wizard and the generated reports vl To Edit the Charts Settings Settings Options Note Chart Colors modify the charts colors Default Chart Bars Orientation horizontal It is possible to set another default vertical orientation for specific report items vV To Edit Report Items 1 Setthe properties of your choice 2 Click Save to validate A window opens indicating that your style sheet has been successfully saved 3 Click OK Properties Functions Displayed as name of the label View Type choose between Tabular HTML and Graphical The last one is only available if the report item can be displayed as a graph Chart Type select one of the proposed chart types Note that this option is only availa
146. d click the option Translate Categories Anew window appears Click the Load button Select the format of the translation in the list Data Type Use the Browse button located on the right of the Fo der field to select the folder or the database in which the description is stored 6 Use the Browse button located on the right of the field Table or File to select the file or the table containing the description 7 Click OK 8 Click the button n Update to refresh the display of the categories 9 Ifthe list of columns is not named correctly use the Advanced Settings a see next paragraph to set a header line and update again 10 Map the language names with those from the loaded translation by clicking the categories and choosing the corresponding language in the contextual menu 11 Click OK CUSTOMER SAP Infinitelnsight 7 0 217 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering selecting Variables Once the training data set and its description have been entered you must select different variables one or more targets variables The Infinitelnsight Modeler Segmentation Clustering feature is capable of segmenting a data set independently that is it does not require that a target variable be selected However even though this is not required we strongly recommend selecting a target variable For the process of segmenting a data set gains maximum meanin
147. d scoring mode can be used if all the following conditions are met the apply in data set table view select statement data manipulation and the results data set are tables coming from the same database the model has been computed while at least one physical key variable was defined in SAP Infinitelnsight there is a valid Infinitelnsight Scorer license for the database no error has occurred the in database apply mode is not deactivated granted access to read and write create table To Use the In database Apply Mode Check the option Use the Direct Apply in the Database and automatically the option Add Score Deviation is selected as well a SAP InfiniteInsight Vx y z2 class_Census01 File Help E Applying the Model Application Data Set Data Type Data Base w Folder kKsrmultiz_sglserwver2005 98 i Se iscccieM Metadata are stored in the same place as data source Define Mapping Generation Options Generate Predicted Value Only wt Advanced Apply Settings Mode Apply w Add Score Deviation Results Generated by the Model Data Type Data Base hal Folder Ksnvmulti2_sglserver2005 at 688 Define Mapping Use Direct Apply in the Database 44 Cancel EJ The input file is missing I Apply CUSTOMER SAP Infinitelnsight 7 0 164 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Clas
148. d side The variable moves to the screen section Excluded Variables Also select a variable in the screen section Excluded Variables and click the button lt to move the variables back to the screen section Explanatory Variables Selected l Note By default any variable defined as a key is put in the Excluded Variables However the user has the possibility to move a key variable in the Explanatory Variables Selected if he wants this variable to have this role 3 Click Next The screen Parameters of the Model appears Checking Modeling Parameters The screen Summary of Modeling Parameters allows you to check the modeling parameters just before generating the model a SAP InfiniteInsight Vx y Z class_Census01 File Help Summary of Modeling Parameters Model Name class Census01 Description Kxen RobustRegression Data to be Modeled _ _ Samples Census Census01 csv Cutting Strategy Random without test Target Variable class Weight Variable Optional None Compute Decision Tree Enable Auto selection jw Export KxShell Script 44 Cancel 41 Previous I Generate l Note The screen Summary of Modeling Parameters contains an Advanced button By clicking this button you access the screen Specific Parameters of the Model For more information about these parameters Setting the Advanced Parameters on page 99 CUSTOMER SAP Infinitelnsight 7 0 96 2014 SAP AG or an SAP affiliate company Al
149. data should be displayed Some reports can be displayed as a pie chart Some reports can be displayed as a line chart When the current report is displayed as a bar chart this option allows you to change the orientation of the bars from horizontal to vertical and vice versa This option allows you to display the current report with no sorting This option allows you to sort the current report by ascending values This option allows you to sort the current report by descending values This option allows you to sort the current report by ascending names This option allows you to sort the current report by descending names Usage Options text editor a spreadsheet a word processing software This option allows you to copy the data from the current view of the displayed report The data can then be pasted in a This option allows you to print the current view of the selected report depending on the chosen display mode HTML y table graph CUSTOMER This option allows you to save under different formats text html pdf rtf the data from the current view of the selected report This option allows you to save under different formats text html pdf rtf the data from all the views of the selected report This option allows you to export to Excel This option allows you to save all reports This option allows you to save the customized style sheet SAP Infinitelnsight 7 0 262 2014 SAP AG or an SAP
150. del Settings Advanced Control for Deviations CUSTOMER SAP Infinitelnsight 7 0 139 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Possible Variables Exclusion Causes The table below shows the possible variables exclusion causes Overall Exclusions Target Specific Exclusions Overall Exclusions Overall Exclusions Target Specific Exclusions Target Specific Exclusions Target Specific Exclusions Target Specific Exclusions Target Specific Exclusions Display Options column Name Constant Small Variance Fully Compressed Small KI on Estimation Small KI on Validation Large KI Difference Small KR Explanation The variable has only one value continuous variables or one category nominal or ordinal variables in the data set The variable is discarded with respect to all targets For continuous variables the variance is small The variable variation is noise The variable is discarded with respect to all targets The variable has been fully compressed with respect to the target It will be excluded from the model with respect to this target The variable has a small KI on Estimation data set with respect to the target It will be excluded from the model with respect to this target The variable has a small KI on Validation data set with respect to the target It will be excluded from the model with respect to this
151. deling results Once the models have been validated you can apply them to Oneor more specific observations taken from your database Simulation mode Anew complete data set or application data set Batch mode To facilitate deployment and integration of the models the code corresponding to each model can also be generated in the programming language The Infinitelnsight Scorer feature which is responsible for generation of this code is described below The Infinitelnsight Scorer Feature The Infinitelnsight Scorer feature formerly known as KMX generates code in the following languages C XML AWK HTML SQL PMML2 SAS or JAVA corresponding to a model generated by SAP Infinitelnsight In this form the model may be integrated into any application that supports the aforementioned languages The generated codes allow the SAP Infinitelnsight models to be integrated within any given application or software package or to be applied directly to the data without requiring SAP Infinitelnsight environment A Caution Code generation is only available for models using the following features Infinitelnsight Modeler Data Encoding Infinitelnsight Modeler Regression Classification Infinitelnsight Modeler Segmentation Clustering CUSTOMER SAP Infinitelnsight 7 0 14 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight 3 3 Methodological Prerequisites Before modeling your data
152. dels with confidence This document is the primary guide to the two SAP Infinitelnsight features described in the following table The feature Allows you to Example I nfinitel nsight Understand and predict a You work for an automobile manufacturer and wish to send a promotional Modeler phenomenon mailing to your prospects Infinitel nsight Modeler Regression Classif Regression Classification allows you to ication Understand why previous prospects responded to such a mailing Predict the response rate to such a mailing sent to new prospects I nfinitel nsight Describe a data set by Your firm is in the process of bringing products A and B to market Modeler breaking it down into Infinitel nsight Modeler Segmentation Clustering allows you to Segmentation Clu homogeneous data Regroup your customers into several homogeneous groups stering groups or clusters Understand the behavior of each of these groups with respect to products A and B 2 2 Before Beginning 2 2 1Files and Documentation Provided with this Guide sample Data Files The evaluation version and the registered version of SAP Infinitelnsight are supplied with sample data files These files allow you to take your first steps using various features of SAP Infinitelnsight and evaluate them During installation of SAP Infinitelnsight the sample files are registered in the folder c Program Files SAP InfiniteInsight Infinite
153. deration Run section Save the model or generate the source code Save Export section Model Overview The Model Overview screen displays the same information as the training summary Overview Name Name of the model created by default from the target variable name and the data set name Data Set Name of the data set Initial Number Variables Number of explanatory variables used O erected Number of explanatory variables actually used by the resulting model Number of Records Number of records in the data set Building Date Date and time when the model was built Learning time Total learning time Depending on the feature used Kxen RobustRegression Kxen SmartSegmenter Kxen TimeSeries Kxen AssociationRules Kxen EventLog Kxen SequenceCoder Kxen SocialNetwork Engine name CUSTOMER SAP Infinitelnsight 7 0 114 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Modeling Warnings Monotonic Variables Indicates if monotonic variables have been found in the data set that is variables which direction Detected of variation is constant in the reading order of the data in the estimation data set Suspicious Variables This report presents a list of variables that are considered to be suspicious These suspicious Detected variables have a predictive power over 0 9 they are very correlated to the target variable This means these variables pro
154. determined that the two explanatory variables that contributed the most information to explain the target variable were the variables marital status and capital gain g SAP InfiniteInsight VX Yy 2 class_Census01 File Help ce Selecting Contributory Variables ST Oe ew we MT oath eo ae copatess BI 25 088 0900 107 O MMM 3 carton MMS oan oo aes gM a Me O oa oo ra O BT i vonas Jie oae os mee O BMT 4 wfo oos oss 1043 O 8 MET O a E moo 4 ria rae fsa 0 089 0 904 1083 O 8 a fiwot 02 0 088 0 995 1033 O Number of Selected Variables 2 44 Cancel CUSTOMER SAP Infinitelnsight 7 0 183 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 7 Click Yes to move to the screen Selecting Variables a SAP InfiniteInsight VX Yy 2 class_Census01 File Help ce Selecting Variables Explanatory Variables Selected 2 capital gain gt aly g M Alphabetic Sort Target Variables 1 class Alphabetic Sort Weight Variable 0 Excluded Variables 13 workclass fniwagt education education num occupation relationship race g 5 C Alphabetic Sort 41 Previous I gt Next 8 Resume the model configuration from the step Selecting Variables Generating the Source Code of a Model The feature used to generate the source code of a model is Infinitelnsight Scorer For more information on t
155. e Consider the following case Using Infinitelnsight Modeler Regression Classification you have contacted the prospects most likely to be interested in your new financial product and identified the ideal number of prospects to contact out of the entire database meeting the deadlines and within the budget you were allowed see the lt DocK2R gt To improve the rate of return of your campaign senior management asks you to Build a segmentation model of your customers Analyze the characteristics of the identified clusters Define customized communications for each cluster The segmentation model in particular should allow you to distinguish customer clusters by virtue of their propensity to purchase the new high end savings product proposed by your firm You will optimize your understanding of your customers 6 1 3 Your Approach For organizational reasons you want to define five groups of customers or clusters and describe the customer profiles for each of these groups To accomplish this project you will use the sample of 50 000 people who responded to your first test during the previous campaign This file corresponds to the sample file CensusOl csv provided with SAP Infinitelnsight and described in the section Introduction to Sample Files see page 60 6 1 4Your Business Issue In your marketing database you have A list of 1 000 000 prospects A list of 50 000 prospects people selected during the
156. e proportion of missed signals or lost opportunity Because the data are ordered from records predicted least likely to be signals on the left to records most likely to represent signals on the right the slower the rise the more sensitive the model in terms of detecting signals or responders The wizard line turns upward from the x axis at the point corresponding to the proportion of non signals in the validation data set CUSTOMER SAP Infinitelnsight 7 0 48 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Lorenz Bad Lorenz Bad displays the cumulative proportion of true negatives specificity accounted for by the bottom x of model scores Here the faster the rise the lower the frequency of erroneous detection Performance Specificity 0 Bey Ae Oty igh ohh ag a 9 8 O41 hag VEA percentage E Random E Wizard Validation 4 10 3 Density Curves The density curves display the density function of the variable Score in the set of Events Curve Density Good and in the set of Non Events Curve Density Good These curves can also be viewed as the derivate of Lorenz curves the density function is by definition the derivate of the cumulative density function The estimated density function in a bin or interval is equal to Manber of Events inthe Interval Total manber af Events Length of the interval The length of an interval is by definition its upper b
157. e 1 1 084e 2 age 3 339e 1 9 982e 3 age 3 107e 1 6 447 e 3 age 2 105e 1 4 2788 3 age 1 455e 1 3 97 3e 3 age 1 356e 1 9 495e 3 age 3 289e 1 3 025e8 3 age 9 5268 2 3 966e 3 age 1 3128 1 014e 4 age 5 7362 3 3 872e 4 age 1 065e 2 3 872e 4 age 1 065e 2 In the case of a continuous variable the Score Card always includes a number of categories that is higher than in the user defined structure or as given by the parameter band count if no user structure has been set Indeed the encoding of variables for the Score Card adds target curve points to increase the accuracy of coding according to the training data set These points split some existing categories and thus increase the number of categories in the Score Card CUSTOMER SAP Infinitelnsight 7 0 142 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Risk Mode The representation of a model equation is easier to read and to interpret in the Risk Mode due to stepwise encoding for ordinal and continuous variables In the Risk Mode it is easy to define which category has a negative or positive effect on the risk score and consequently on the odds or on the probability of risk In order to illustrate the advantages of a scorecard in interpreting results the variable age will be used for this example The segment 24 27 has a risk score of about 30 and the segment 37 43 h
158. e CensusOl cvs data file vV To Select a Description File 1 Onthe screen Data Description click the button Open Description The following window opens i Load a Description for Census01 csv Folder Samples Census w Fleane ooo Og ok 2 Inthe window Load a Description select the type of your description file 3 Inthe Folder field select the folder where the description file is located with the Browse button l Note The folder selected by default is the same as the one you selected on the screen Data to be Modeled CUSTOMER SAP Infinitelnsight 7 0 70 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 4 Inthe Description field select the file containing the data set description with the Browse button A Caution When the space used for model training contains a physical variable named kxIndex it is not possible to use a description file without any key for the described space When the space used for model training does not contain a physical variable named kxIndex it is not possible to use a description file including a description about a KxIndex variable since it does not exist in current space 5 Click OK The window Load a Description closes and the description is displayed on the screen Data Description a SAP InfiniteInsight Vx y zZ New Regression Classification Model File Help Data Description BH l Open Description
159. e Display Settings Whole Population Population 48842 Positive Target b3 93 i marital status marital status marital status marital status Mewer married Manned AF spouse M Diverved Marnied sp Separated Widowed Population 16717 Population 22416 Population T251 Population 3048 Positive Target 4 55 Positive Target BEE o x Positive Target 10 04 Positive Target 7 45 Estimation Validation Population Count 36381 12461 48642 Positive Target Count a714 2973 11687 Positive Target Ratio 23 95 23 86 23 93 Negative Target Count 2 667 94868 37155 Negative Target Ratio 76 05 76 14 TEOT Variance 0 19 0 02 Weighted Population 36381 0 12461 0 CUSTOMER SAP Infinitelnsight 7 0 147 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Understanding the Decision Tree The panel Decision Tree is split into three parts 1 the decision tree itself which is displayed in the upper section of the panel 2 twotabs located in the left bottom part of the panel provide you with information on the nodes and with the profit curve corresponding to the current decision tree 3 anavigator allowing you to visualize what part of the tree you are studying Is displayed in the right bottom part of the panel 4 SAP InfiniteInsight VX y 2 cass_Census01 File Help ali Decision Tree Target E ass ae Save asImage Display Setti
160. e E 29 4 6 5 ROES Ol VAS e aa a a 30 KAE o E EE E eee E E E E T 35 4 7 1 FUNC Ae Talia he NN UO ee a a aa 35 4 7 2 Ferro nance ora Model eea rts Ore ah dae ah eed ahaa a Seed raa iare 35 4 7 3 VOSS OR Mode ran a a A a E et Lees sedate sedate sada el bala eas 2oa8 3 bac 35 4 7 4 ASTICP AUIS a Modelhane a a 36 4 7 5 Representation Ora ModE ke E E EA E EAEE E EEEE E AAA 36 4 7 6 Mandatin Tine MOCE eraann enee EE NNNNA 37 4 7 7 Under what Circumstances is a Model Acceptable ssssssssssennirrrrrrrrrssssrrrerrrrrrrrrrrrrrrrsrsrrrrrrrrrrn 38 4 7 8 Howto Obtaina Better Modelteoens tienenie a a a a a E E EIES 38 CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved How to Use this Document 4 8 4 9 4 10 BZ 6 2 Fe TSO MC NCL ALO et Bacchi tee Ne on cake ala dual E N EE EEN EE EE AEEA EEE AEAEE E T 39 4 8 1 imdicators Specifice to SAP Inimitelnsie M riea a a a a Ol A 39 4 8 2 Other conimony Used Mde easain a N 41 4 8 3 EPORA CH OLSE tre ee rev E NN 43 PEOI EE E E E AN 46 4 9 1 Yea ig lie 6 1g leery rer erry ree ee aE EEEE E E reer 46 4 9 2 Avala DIE FTON I ype Sas a NEEE E 46 Advanced Model O10 a1 oxo eee ee TS 47 MNO ase ages tac tec are ea eats age ner nan nian aetna clones ten ce tate ie sea meta ee MN etal ately 47 AI Eren O UNES Se enn nr PR ee OE PO PAM Oar ne ee Oe aetna 48 AIOS DEn YOUVE S NN EEEE E a 49 AAA R ce ENO A AO AEA T OTT a a a aa a a en Te 5 Infinitelnsight Mode
161. e Help Simulating the Model Explanatory Variables Sort by Contribution of class ha Er oo capital gain occupation education num age capital loss 3 Variable marital status hours per week Min education Max relationship sex workclass native country E E E G2 Ga Ga T Ga Ga G2 Ga Ea EEE Tace Results co CUSTOMER SAP Infinitelnsight 7 0 178 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 3 172 4 SAP InfiniteInsight VX y 2 class_Census01 File Help Explanatory Variables Sort by Contribution of class cCapital gain occupation Divorced education num Married AF spouse Married civ spouse age capitalloss Married spouse absent hours per week Never married education Separated Widowed relationship sex workclass native country race EEE Results CUSTOMER amp Cid Ga D E a Ga Gy E a E Variable marital status Min Max In the section Modifying values in the Value field select or enter a value such as Married civ spouse The value appears in the table of Explanatory variables across from the selected variable SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 4 Ifyou would like to select other explanatory variables
162. e button gt located on the left of the screen section Weight Variable middle right hand side The variable moves to the screen section Weight Variable Also select a variable in the screen section Weight Variable and click the button lt to move the variables back to the screen section Explanatory variables selected CUSTOMER SAP Infinitelnsight 7 0 94 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Explanatory Variables By default and with the exception of key variables such as Kx ndex all variables contained in your data set are taken into consideration for generation of the model You may exclude some of these variables For the first analysis of your data set we recommend that you retain all variables It is particularly important to retain even the variables that seem to have no impact on the target variable If indeed these variables have no impact on the target variable the model will confirm this In the contrary case the model will allow you to recognize previously unidentified correlations between these variables and the target variable By excluding variables from the analysis based on simple intuition you take the risk of depriving yourself of one of the greatest value added features of SAP Infinitelnsight models the discovery of non intuitive information Depending on the results obtained from the first analysis which included all of the variables of
163. e distributed as a block to the test data Set CUSTOMER SAP Infinitelnsight 7 0 23 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Sequential Without Test The Sequential without test strategy cuts the initial data set into two blocks The lines corresponding to the first 3 4 of the initial data set are distributed as a block to the estimation data set The lines corresponding to the next 1 4 of the initial data set are distributed as a block to the validation data set As no test sub set is used all the data from your training data set can be used for sub sets of estimation and validation This can lead to a model with a better quality and robustness 4 5 Table of Data 4 5 1Definition A table of data is a data set presented in the form of a two dimensional table In this table Each row represents an observation to be processed such as American individual in the sample file CensuSsOl1 csv Each column represents a variable that describes observations such as the age or the gender of individual Americans Each cell the intersection of a column and a row represents the value of the variable in the column for the observation in that row The following table is an example of a table of data Observations Variable 1 Variable 2 Variable 3 Observation a Value al Value a2 Value a3 Observation b Value b1 Value b2 Value b3 Observation n Value n1 Value n2 Value n3 CUSTOMER
164. e encoding of variables adds target curve points to increase the accuracy of coding according to the training data set These points split some existing categories and thus increase the number of categories in the generated code CUSTOMER SAP Infinitelnsight 7 0 185 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 3 In the section Code Settings select the code type to be generated List of Generated Codes on page 188 4 Click the Browse button associated with the Folder field and select a folder to save the generated file In the field Generated File enter the name of the exported file If you want to replace an existing file use 5 186 the Browse button to select it Select Source Folder for Data SEM Samples H E Census BG JapaneseData GT KAR Gl KelData GG KSN AKTE GI KTs a g CA a g J H KA H PA a g T H Ti gi U Samples D w Tal Text Files dat data csv bet hl Use r Passwa rd CUSTOMER oK Cancel SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 6 If you have selected the option View Generated Code it is displayed at the end of the generation Process 7 Click the Generate button The figure below shows the beginning of a sample C source code of an Infinitelnsight m
165. e field First Row Index enter the number of the first row you want to display 3 Inthe field Last 4 Click the Refresh button to see the selected rows Row Index enter the number of the last row you want to display A Comment about Database Keys For data and performance management purposes the data set to be analyzed must contain a variable that serves as a key variable Two cases should be considered If the initial data set does not contain a key variable a variable index Kx ndex is automatically generated by SAP Infinitelnsight features This will correspond to the row number of the processed data If the file contains one or more key variables they are not recognized automatically You must specify them manually in the data description See the procedure To Specify that a Variable is a Key see page 75 CUSTOMER SAP Infinitelnsight 7 0 74 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification To Specify that a Variable is a Key 1 Inthe Key column click the box corresponding to the row of the key variable 2 Type inthe value 1 to define this as a key variable CUSTOMER SAP Infinitelnsight 7 0 75 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Defining a Variable Structure There are three ways to define a variable structure by first extracting the categor
166. e usable by SAP Infinitelnsight features the data set to be analyzed must be presented in the form of a single table of data on page 24 except in instances where you are using the Event Logging or Sequence Coding features of Infinitelnsight Explorer To use SAP Infinitelnsight features you must have a training data set available that contains the target variable with all its values defined Then you can apply the model generated using the training data set to one or more application data sets The training data set is cut into three data sub sets for estimation validation and testing using a cutting strategy on page 19 The different types of variables on page 26 continuous ordinal and nominal are next encoded by the Data Encoding feature of Infinitelnsight Modeler or by the Event Logging and Sequence Coding features in the case of dynamic data Before generating the model you must Describe the data A utility integrated with SAP Infinitelnsight allows you to generate a description of the data set to be analyzed automatically You need only validate that description verifying that the type and storage format of each variable were identified correctly Define the role of variables contained in the data set to be analyzed You may select one or more variables as target variables These are the variables that corresponds with your business issue The other variables of the table of data are considered to be explanat
167. each customer will invariably grow with time The more data your database accumulates the harder it is for you to manually create clusters that take all data into consideration and to develop a response to your business issue Furthermore as the increasing volume of information requires you to build segmentation models with increasing frequency the time required to build these segmentation models becomes increasingly more significant Finally management may want you to rationalize your methods and to perform your segmentation using a method not based purely on your intuition Defending your segmentation method based on intuition may be difficult CUSTOMER SAP Infinitelnsight 7 0 199 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Classical Statistical Method On the basis of the information that you have a Data Mining expert could build a segmentation model In other words you could ask a statistical expert to create a mathematical model that would allow you to build clusters based on the profiles of your customers To implement this method the statistician must Perform a detailed analysis of your database Prepare your database down to the smallest detail specifically encoding the variables as a function of their type nominal ordinal or continuous in preparation for segmentation The encoding strategy used will determine the type of segmentation model obta
168. eatures require only an extremely short modeling time to generate relevant and robust analytical models of your data Infinitelnsight Modeler Regression Classification formerly known as K2R generates explanatory and predictive models The models generated by Classification Regression explain and predict a phenomenon or business question by a function of the analyzed data set the explanatory variables The models generated are calculated using a regression and classification algorithm This polynomial regression is a proprietary algorithm using Vapnik s SRM Structural Risk Minimization principle to calculate the parameters Infinitelnsight Modeler Segmentation Clustering formerly known as K2S generates descriptive models which means a function to regroup cases in a data set into a number of clusters with similar behavior toward a business question Infinitelnsight Modeler Time Series formerly known as KTS lets you build predictive models from data representing time series Thanks to Infinitelnsight Modeler Time Series models you can Identify and understand the phenomenon represented by your time series Forecast the evolution of time series in the short and medium term that is predict their future values Phase 4 Model Presentation and Deployment Once the models have been generated model performance indicators plots and modeling reports in HTML format facilitate viewing and interpretation of the data mo
169. ecessary for accessing the data database A database is a structured collection of records or data that is stored in a computer system descriptive model A model which allows describing data sets CUSTOMER SAP Infinitelnsight 7 0 276 2014 SAP AG or an SAP affiliate company All rights reserved Glossary detected profit Detected profit is the profit type shown as the default It allows examination of the percentage of observations belonging to the target category of the target variable that is the least frequent category as a function of the proportion of observations selected from the entire data set determination coefficient R2 ratio between the variability sum of squares of the prediction and the variability sum of squares of the data deviation Deviation is a measure of difference for interval and ratio variables between the observed value and the mean domain See Analytical Record The behavioral domain is usually obtained through aggregates per entity on transactional tables E encoding Encoding is the process of putting a sequence of characters letters numbers punctuation and certain symbols into a specialized format for efficient transmission or storage engine The Ul and view independent portion of an application concerned with data manipulation and other fundamental operations independently of how these are eventually represented to the user CUSTOMER SAP Infinitelnsight 7 0 277 2
170. ed by numbers The variable eye color is a nominal variable The set of values that this variable may assume blue brown black for example are clearly distinct non ordered categories and are represented by character strings Nominal Variables and Modeling During modeling the values of the categorical variables are regrouped into homogeneous categories These categories are then ordered as a function of their relative contribution with respect to the values of the target variable CUSTOMER SAP Infinitelnsight 7 0 28 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 6 4 Storage Formats To describe the data SAP Infinitelnsight uses four types of storage formats date datetime number integer string The following table describes these storage formats The storage format date datetime number integer string l Note Is used to describe variables when their values correspond to Dates expressed in the following formats YYYY MM DD YYYY MM DD Dates and times expressed in the following formats YYYY MM DD HH MM SS YYYY MM DD HH MM SS Figures or numerical values on which operations may be performed Figures or numerical integer values on which operations may be performed Alphanumeric character strings For instance 2001 11 30 1999 04 28 2001 11 30 14 08 17 1999 04 28 07 21 58 The variab
171. edicted values generated by the model The result file obtained contains a column in which a category of the target variable is assigned to every observation The decision is taken on the basis of a threshold that is applied on the scores generated by the model The target category of the target variable is assigned to observations whose scores are superior to the threshold The default threshold computed during the generation or training of the model is chosen so that the way the categories of the target variable are assigned to observations is representative from their distribution in the training data set Upon the level of information desired you can choose to generate among several results files described in the table below Selecting the option Predicted value only Probability Individual Contributions Decision For this Scenario Will generate a results file containing the following information Only the predicted value of observations rr_ TargetVariableName the predicted value the probability oroba_rr_TargetVariableName the prediction range bar_rr_TargetVariableName the predicted value the probability the prediction range the individual contributions of variables contrib_VariableName_rr_TargetVariableName the predicted value the decision decision_rr_lTargetVariableName the decision probability Oroba_decision_rr_lTargetVariableName the probability Due
172. ee sub sets estimation validation and test Random with Test at the End The Random with test at the end cutting strategy distributes 4 5 of the initial data set in a random manner in the two sub sets estimation and validation 3 5 being distributed in the estimation data sub set and 1 5 in the validation data sub set The final 1 5 of the initial data set is sent directly into the test sub set This is a useful strategy in cases where Your database corresponds to a well defined evolution because of the way it was built which may mean for example that the data Is in chronological order You may wish to take this order into account when generating your model For example imagine that New customers are added every month to your database You know that the data sets to which you apply the model will once generated have a better chance of resembling the most recent section of your database that is the section that contains the most recent customers entered Using the Random with test at the end cutting strategy you decide to test the model generated on that section of your database that is most likely to resemble the state of your future application data sets Random Without Test Default Strategy The Random without test strategy is the cutting strategy suggested as the default setting It distributes the whole initial data set in a random manner to the two sub sets of estimation and validation 3 4 of t
173. enerated using SAP Infinitelnsight This indicator corresponds to the proportion of information contained in the target variable that the explanatory variables are able to explain Example A model with a predictive power of 0 79 is capable of explaining 79 of the information contained in the target variable using the explanatory variables contained in the data set analyzed lL is a hypothetical perfect model capable of explaining 100 of the target variable using the explanatory variables contained in the data set analyzed In practice such a predictive power would generally indicate that an explanatory variable 100 correlated with the target variable was not excluded from the data set analyzed OQ is a purely random model Improving the Predictive Power of a Model To improve the predictive power of a model new variables may be added to the training data set Explanatory variables may also be combined CUSTOMER SAP Infinitelnsight 7 0 39 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Robustness Indicator Prediction Confidence Definition The prediction confidence is the robustness indicator of the models generated using SAP Infinitelnsight It indicates the capacity of the model to achieve the same performance when it is applied to a new data set exhibiting the same characteristics as the training data set Example A model with a prediction confidence Equal to o
174. er observation having a higher score than a non signal non responder observation For individual variables ordering based on score is replaced by ordering based on the response probability for the variable s categories for example cluster ID or age range response rates authenticated server Users will be able to communicate to SAP Infinitelnsight authenticated server only when providing correct password SAP Infinitelnsight authenticated server delegates the authentication to Custom built services or Operating System services through PAM Pluggable Authentication Modules autoselection It is an automated attribute selection CUSTOMER SAP Infinitelnsight 7 0 272 2014 SAP AG or an SAP affiliate company All rights reserved Glossary bin A bin is a range of values defined by its bounds upper bound and lower bound Bins result from a data manipulation activity known as binning Synonym range bipartite graph display non bipartite graph display The bipartite graph display shows two distinct populations of nodes or node sets with the links between the two node sets For example the first node set could represent clients and the second products From this global view a non bipartite graph display can be derived to focus on the links between the nodes of a given node set bubble chart A bubble chart is a specific graphical representation in Infinitelnsight Modeler Segmentation Clustering which displays clusters a
175. erate a KxShell script reproducing the current model This script can be used to run models in batches One easy way to get special settings in exported KxShell scripts is to first do the corresponding operation in the graphical user interface For example if you run an auto selection of variables before exporting the Shell script then the exported script will include the code needed to do the auto reduction vV To Save the KxShell Script 1 Inthe section Save Export of the menu Using the Model select the option Export KxShell Script The panel KxShell Script Generation is displayed a SAP InfiniteInsight VX y 2 class_Census01 File Help 3 KxShell Script Generation Kx Shell Script Saving Location Folder USamples Census Model Data Set Description Saving a Save the Descriptions in the Script Save the Descriptions with the Script Save the Descriptions with the Data Save the Descriptions Separately a Learn Apply Generate Variable Structure From Statistics Script Preview Select a Target 2 Use the Browse button located to the right of the Folder field to select where the script will be saved 3 Inthe field KxShel Script enter the name of the file in which the script will be saved 4 Inthe frame Model Data Set Description Saving select where you want to save the data description The four available options are Save the Description in the Script the data description is added in the KxShell scr
176. erequisites 5 855 inciniac intintadinhant net aiohsd nadie daabdanbdinadbsetinbdimbtn adie Gu GiGu Geass ltaeeetta eats 15 3 1 WV VAIS your BUISINCSS 1S SUC annan a k hs Rag ct a at ce gis aes 15 3 3 2 SOIT ea Sa a cre ti cela cuca de cua desde de ah dea dea ead demo ardeweduasdeosdenndemedeonteanteaetaavaees 15 Essential CONCEDUS wciavonsiutavenudnaincancnansnawananarewsbavenswacsasusionaisnscnuan sabi A a Ea E 16 Op ratin o SAP NANmtensie Nt Overview iatna a E a ae E ck Ei A hae he 17 BA OUr ES o PDO E ian ia sede siae eoniden luce daalsennistneniechicg ce chaas eshte ehst cece ceacsaead da dad 18 Dira es eg te ee te ei ar E E svn TOP A EE Pn eT PE POO AS oe Se ee 18 4 3 1 TANN Be ters eo C Essee E a a a a 18 4 3 2 Application Data Seti inace i l DP 19 GE o eop ES e a a A a a A 19 4 4 1 BiS ES AEEA A E A E A A AAAS AASS AA AAA AAA AAAA AAA AAA EAA EEEE EEA 19 4 4 2 BAO evoke MH ale Ra eea E Re en E O E E E E E AAE eee 20 4 4 3 Type ge GC eral a freee aeoe yra O E E T R 20 TABE CA ass as A a ee a as E E EA ee one 24 4 5 1 E E e a A E a a O es 24 4 5 2 Synonyms of Observation and Variable css aa a 5 Oe 25 4 5 3 AT AP ON MNS e a seaeaene cat scatateb stonatabatarateb canta veiaistaiaias4 25 Wie VAD OS teases as cue a E ate ote rete acest E eae eed etc ce AE MAE as 25 4 6 1 Generne De TIO eii 25 4 6 2 hy 6 ciie nn rT re 26 4 6 3 i a1 49 chor 1H aaO eee ene een eee ee a aa eee eee eee eee 26 4 6 4 Storage ax oO Fo hae en a Pe
177. erformance Chart 1 600 A A 1 0 1 500 A i 1 400 0 8 1 300 1 0 8 1 200 w 1 100 0 7 a 1 000 j 0 6 D 900 y g s00 i Foo ol a soo 7 0 0 af 2 F 6 ane a ah ae ae ee en ae eb ofr el ee ae at aah af Fr a gi ga 40 score E Density Validation Odds Validation Probability Validation CUSTOMER 33 SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 5 Infinitelnsight Modeler Regression Classification IN THIS CHAPTER Application Scenario Enhance Efficiency and Master your Budget using Modeling cssesseeeeeeeeeeeeeeeeeeees 54 Creating a Classification Model Using Infinitelnsight Modeler cccccccccceeeeeeeeceeeeeeeeeeeeeeeeeeeeesseeeeeeeesaneaas 66 5 1 Application Scenario Enhance Efficiency and Master your Budget using Modeling 5 1 1 Presentation In this scenario you are the Marketing Director of a large retail bank The bank wants to offer a new financial product to its customers Your project consists of launching a direct marketing campaign aimed at promoting this product You have a large database of prospects at your disposal and a limited and closely monitored budget and you are also subject to significant time constraints In order to maximize the benefits of your Campaign your business Issue consists of Contacting those prospects most li
178. error bar In other words the error bar is a deviation measure of the values around the predicted score It appears in the output file as out lier_rr_ lt target variable gt Possible values are 1 if the observation is an outlier with respect to the current target else O CUSTOMER SAP Infinitelnsight 7 0 169 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Predicted Value Quantile This option allows you to cut the output file in quantiles and to assign to each observation the number of the quantile containing it Approximate quantiles are constructed based on the sorted distribution and the boundaries of predicted scores from the validation sample The score boundaries are used to determine approximate quantiles on the apply data set l Note Exact quantile computation would require a full sort of the scores obtained on the apply data set which can be consuming From version V6 0 SAP Infinitelnsight offers Gain Chart option for this purpose It appears in the output file as quantile_rr_ lt target variable gt _ lt number of quantiles gt for example for a target variable named class and a number of quantiles equal to 10 the generated column will be named quantile_rr_class_10 1 Check the option Predicted Value Quantiles 2 Inthe field Number of Quantiles enter the number of quantiles you want to create Check the option Predicted Value Quantiles Co
179. ers For instance it can be useful during a presentation The X axis the Y axis and the bubble size represent one variable each You can define the variables to use in the chart Thus you can create a bubble chart that separates distinctly the clusters from each other enabling you to identify the clusters of interest for your marketing campaign The figure below represents the relationship between the variables frequency class and capital gain For instance results show that the customers listed in cluster 6 are earning 10 163 4 dollars per year in average capital gain 10 163 4 and represents 5 5 Frequency 0 055 of the population listed in the data set In addition among these customers 85 5 class 0 885 responded in a positive manner to the test phase of your marketing campaign In comparison cluster 2 represents the biggest population listed in the data set namely 25 2 of the population Frequency 0 252 which is around five time bigger than the population listed in cluster 6 However the customers listed in this cluster are earning less the customers listed in cluster 6 147 542 dollars per year in average capital gain 147 542 thus 70 less than cluster 6 Moreover among the customers listed in cluster 2 only 27 16 class 27 16 responded in a positive manner to the test phase of your marketing campaign Consequently compared to cluster 2 cluster 6 is more interesting because it showed better results to the test
180. es Among the five clusters Cluster 2 is the one which contains the greatest number of observations or 25 2 of the total number of customers contained in the entire data set The Plot Relative Target Means Similar to the Target Means plot the Relative Target Means plot presents the proportion of observations for each cluster belonging to the target category of the target variable The only difference between the two plots is the scale used on the Y axis On the Relative Target Means plot the proportion of observations belonging to the target category of the target variable relative to the entire data set is re expressed In other words the O value of the Y axis corresponds to the true percentage of observations belonging to the target category of the target variable in relation to the entire data set The figure below presents the Relative Target Means plot obtained during this scenario The bars have been sorted in descending order Relative Target Means Data Set Estimation 6 0 6167791485 7804358 a ja g E D 1 an i Clusters CUSTOMER SAP Infinitelnsight 7 0 251 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Among the ten clusters Cluster 6 is the cluster that has the highest proportion of observations belonging to the target category of the target variable Compared to the entire data set Cluster 6 contains 61 6 more customers belonging to t
181. es and actual target values Infinitelnsight technology limits this effect by providing piece wise linear recalibration of the estimates to the actual targets based on the statistics on the Validation data set thus providing not only good order estimates but also good range estimates 4 8 2 Other Commonly Used Indicators Three other indicators commonly used in Data Mining are provided to assess a SAP Infinitelnsight model the GINI index the K S the AUC GINI Index The Gini statistic is a measure of predictive power based on the Lorenz curve Itis proportionate to the area between the random line and the Model curve The GINI index is defined as the area under the Lorenz curve see page 48 The GINI index is the area between the Trade off curve and the obtained curve multiplied by 2 This is often pictured as the following chart 1 0 8 aG Sensitivity 0 4 0 2 0 0 02 O04 O06 O18 1 CUSTOMER SAP Infinitelnsight 7 0 41 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts The horizontal axis grows with the score and can be associated with 1 f This is simply expressed using our notations as l Ho GINI A 1 ate ape AAA Pg a CINI 20 Pe 5 1 auc fl pe 2 AUC T Using these notations we know that the GINI index of a random model is O and for a perfect modelis 1 Pe K S K S is the Kolmogorov Smirnov statistic applied here as a meas
182. es not come from the typical population of data in other words extreme values Ina normal distribution outliers are typically at least 3 standard deviations from the mean p performance indicator PI Performance indicators help organizations achieve organizational goals through the definition and measurement of progress The purpose of defining Pls is to have a common definition of a metric across multiple projects A metric like customer value could easily be defined in several different ways leading to confusing or contradictory results from one analysis to the next Shared Pls ensure consistency across analysts and projects over time The key indicators are agreed upon by an organization and are indicators which can be measured and will reflect success factors The Pls selected must reflect the organization s goals they must be the key to its success and they must be measurable CUSTOMER SAP Infinitelnsight 7 0 284 2014 SAP AG or an SAP affiliate company All rights reserved Glossary periodic cutting strategy The periodic cutting strategy is implemented by following this distribution cycle 1 Three lines of the initial data set are distributed to the estimation sub set One line is distributed to the validation sub set 2 3 Oneline is distributed to the test sub set 4 Distribution begins again at step 1 pivot A pivot is a data summarization tool found in data visualization programs Among other functions they can aut
183. es of each variable with respect to the target variable Presentation of the SAP Infinitelnsight User Menu Once the model has been generated Click Next The screen Using the Model appears a SAP InfiniteInsight Vx y Z class_Census01 File Help Q Using the Model ai Display Run Model Overview Model Graphs Save Export Contributions by Variables Category Significance Statistical Reports Scorecard Confusion Matrix 44 Cancel l Previous CUSTOMER SAP Infinitelnsight 7 0 113 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification The screen Using the Model presents the various options for using a model that allow you to Display the information relating to the model just generated or opened Display section referring to the model curve plots contributions by variables the various variables themselves HTML statistical reports table debriefing Some information is only displayed upon request from the user the display of Infinitelnsight Modeler Regression Classification results as a decision tree which can be specified in the modeling parameters before the model generation or the display of model parameters which can be requested in the general user options Apply the model just generated or opened to new data to run simulations and to refine the model by performing automatic selection of the explanatory variables to be taken into consi
184. eserved Infinitelnsight Modeler Segmentation Clustering 2 Check the option Enable Model Autosave A SAP InfiniteInsight VX y 2 cass_Census01 File Help Model Autosave Enable Model Autosave Description Data Type Tex Files kal Folder File Table E Browse i Samples Census wv E Browse EJ You must select a filefable 4l Previous I gt OK 3 Setthe parameters listed in the following table Parameter Model Name Description Data Type Folder File Table 4 Click OK CUSTOMER Description This field allows you to associate a name with the model This name will then appear in the list of models to be offered when you open an existing model This field allows you to enter the information you want such as the name of the training data set used the polynomial degree or the performance indicators obtained This information could be useful to you later for identifying your model Note that this description will be used instead of the one entered in the panel Summary of Modeling Parameters this list allows you to select the type of storage in which you want to save your model The following options are available Text files to save the model in a text file Database to save the model in a database Flat Memory to save the model in the active memory SAS Files to save the model in a SAS compatible file for a specified version of SAS and a specified platform
185. essions using a post processing consists in using first an encoded target value instead of the original y target value during the learning model phase in order to have a uniform distribution itisthe pre processing phase Then regression la coefficients are computed and scores are transformed in the original target space during the post processing phase l Note This strategy is to be preferred when the default strategy does not produce models with enough quality which is often the case with very skewed target distributions Example of Performance Curve Performance Performance Performance SAP Infinitelnsight 7 0 102 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification vV Regression Without Post processing Uncheck the option Enable Post processing _ Enable post processing l Note Itis not possible to change the target encoding strategy when the post processing is disabled vV Regression with Original Target Values 1 Check the option Enable Post processing 2 Select the radio button Original target encoding Enable post processing Original target encoding Uniform target encoding l Note This regression strategy corresponds to regressions used in SAP Infinitelnsight from version 3 3 1 to version 3 3 6 This strategy is set by default vV Regression with Uniform Target Encoding 1 Check the option Enable Post processing
186. est scores Popul ive Te heh oR eh oh oha o O AB 40 46 aD a6 1 CUSTOMER SAP Infinitelnsight 7 0 152 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification On the contrary if you expand the node with the lowest percentage of positive target the profit curve will improve over the last percentiles However if the node you expand contains a very small population the profit curve will not be impacted So you need to find the best compromise between the size of the population and the percentage of positive target Customizing the Display The button Display Settings allows you to customize some of the display settings for the decision tree CUSTOMER SAP Infinitelnsight 7 0 153 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Orientation this setting allows you to select if you want to display the tree horizontally or vertically Horizontal Vertical Display Type this setting allows you to display the decision tree as a standard decision tree Decision Tree Like or with a specific Infinitelnsight loOK Infinitelnsight Display The option Decision Tree Like is more compact but the nfinitelnsight Display is more easily read Infinitelnsight Display Decision Tree Like miata stats lives bes iad oe ed ae When you have set the display parameters
187. eters of the Model click the Generate button The screen raining the Model appears The model is being generated A progress bar allows you to follow the process if SAP InfiniteInsight Vx y z class_Census01 File Help Training the Model E mee E Stop View Type Copy Print Save Export to PowerPoint Starting Model learning Stop Current Task 44 Cancel 4 Previous i gt Next 2 Ifthe Autosave option has been activated in the panel Summary of Modeling Parameters a message is displayed at the end of the learning process confirming that the model has been saved i SAP InfiniteInsight Messages ieee Log mx Copy Clear Information New 5 6 14 9 25 07 AM The model class _Census01 has been saved Close CUSTOMER SAP Infinitelnsight 7 0 227 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 3 Click Close 4 Once the model has been generated click Next to go to panel Using the Model Following the Progress of the Generation Process There is two ways for you to follow the progress of the generation process The Progress Bar displays the progression for each step of the process It is the screen displayed by default The Detailed Log displays the details of each step of the process vV To display the Progression Bar Click View Type and select the Progression button The progression bar screen appe
188. ferring to the model curve plots plotting of clusters contributions by variables and the profiles of variables of each cluster Apply the model generated to new data Run section Save the model or generate the source code Save Export section CUSTOMER SAP Infinitelnsight 7 0 232 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Model Overview The screen Model Overview displays the same information as the training summary Overview Model Data Set Initial Number of Variables Number of Selected Variables Number of Records Building Date Learning Time Engine name Requested Number of Clusters SQL Expressions name of the model created from the target variable name and the data set name name of the data set number of variables in the data set number of explanatory variables used to build the model number of records in the data set date and time when the model was built total learning time name of the feature used to build the model Kxen KMeans for a segmentation number of clusters that have been asked for by the user indicates if the SQL expressions for the clusters definitions have been calculated Enabled or not Disabled Nominal Target Variables For each nominal target lt TargetVariableName gt Target Key lt NonTargetCategory gt Frequency lt TargetCategory gt Frequency CUSTOMER name of the target variable
189. for non pathological continuous targets that is continuous targets without distribution peak Dirac as Z P S gt median 5 1 P S gt median 5 In most cases a good approximation is 0 25 CUSTOMER SAP Infinitelnsight 7 0 135 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Normal Profit Properties There are several interesting things to note about normal profit 1 The normal profit of category is independent of the target values themselves user can change the target value through monotonic transformations the normal profit of the categories with respect to this target will not change This belongs to non parametric metrics 2 Aconsequence of 1 is that this metric is resistant to outliers when there are a few occurrences of the target with very high values with respect to the rest of the target value distributions the notion of normal profit is not impacted 3 The weighted sum of the normal profit for all categories of a given variables will always be O Grouping Categories On the plot of details of a variable categories may appear grouped When the option Optimal Grouping is enabled SAP Infinitelnsight groups those categories sharing the same effect on the target variable For instance for the variable education the categories Doctorate and Prof School are grouped If the explanatory variable is continuous SAP Infinitelnsight
190. for which the statistics are displayed wanted target value frequency in percentage of the non target value in the entire data set frequency in percentage of the wanted target value in the entire data set SAP Infinitelnsight 7 0 233 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Continuous Targets Number For each continuous target lt TargetVariableName name of the target variable for which the statistics are displayed gt Min minimum value for the target Max maximum value for the target Mean mean of the target Standard Deviation mean of the distance between the target values and the Mean Performance Indicators For each target variable Predictive Power K I For more information on the predictive power see section Performance indicators on page 39 Prediction Confidence KR For more information on the prediction confidence see section Performance indicators on page 39 Clusters Counts For each target variable Requested Number of Clusters number of clusters that have been asked for by the user Effective Number of Clusters number of clusters found by the model Model Overview Options VI To Copy the Model Overview Click the Copy button The application copies the HTML code of the screen You can paste into a word processing or spreadsheet program a text editor CUSTOMER SAP Infinitelnsight 7 0 234 2014 SAP AG
191. g campaign into homogeneous groups see Summary of the Infinitelnsight Modeler Regression Classification Application Scenario Describe each of these groups and provide customized communication for each of these different groups The Infinitelnsight Modeler Segmentation Clustering feature allows you to create descriptive models The first step in the modeling process consists of defining the modeling parameters Select a data source to be used as a training data set Describe the selected data set on page 210 Select the variables Select the explanatory variables Checking the Modeling Parameters Define the number of clusters This step is optional O oF Q NY a CUSTOMER SAP Infinitelnsight 7 0 207 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering selecting a Data Source For this Tutorial Use the file CensusOl csv as a training data set This file represents the sample that you had extracted from your database and used for the test phase of your direct marketing campaign As specified in your test plan this file contains 50 000 prospects of which you know the behavior with respect to the new financial product 25 of the prospects showed themselves to be clearly interested They chose to accept a meeting with one of your sales agents 75 of the prospects declined your invitation In this file you created a new variable Class which corres
192. g only when it is accomplished in relation to a domain specific business issue expressed in the form of a target variable possibly a weight variable and the explanatory variables l Note For more information on variable roles see section Role of Variables see Roles of Variables on page 30 CUSTOMER SAP Infinitelnsight 7 0 218 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Target Variables For this Scenario Select the variable Class as your target variable that is the variable that indicates the probability of an individual responding in a positive or negative manner to your campaign To Select Targets Variables 1 Onthe screen Selecting Variables in the section Explanatory variables selected left hand side select the variables you want to use as Target Variables fe SAP InfiniteInsight Vx y z New Regression Classification Model File Help Selecting Variables Explanatory Variables Selected 14 Target Variables 1 workclass fniwgt education education num marital status occupation relationship og face SEX _ Alphabetic Sort capital gain Weight Variable 0 capitalloss hours per week ane Excluded Variables ff Aw H _ Alphabetic Sort M Alphabetic Sort l Note On the screen Selecting Variables variables are presented in the same order as that in which they appear in the table of data To sort them alphabeti
193. gmentation Clustering automatically detects interactions between the variables to build homogeneous sub sets or clusters Each cluster is homogeneous with respect to the entire set of variables and in particular with respect to the target variable that is for example responded positively to my test You will discover the characteristics of different clusters such as those clusters with an excellent response rate and those with a poor response rate In addition if your customer database contains customer expenditures on your other products you will also obtain information on product sale synergies by cluster Using Infinitelnsight Modeler Segmentation Clustering you have access to all the analytical features needed to define the type of message to be sent to the cluster for each customer You have homogeneous clusters that will allow you to respond to your business issue Of particular importance this segmentation is systematic the results obtained do not represent a particular point of view of your data and is robust or consistent Two people performing this segmentation using the Infinitelnsight method would obtain the same results CUSTOMER SAP Infinitelnsight 7 0 200 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 6 1 6Introduction to Sample Files This guide is accompanied by the following sample data files A data file CensusOl csv The corresponding de
194. go back to step 2 Otherwise go to step 5 5 Click the Run button to perform a model simulation The results of the simulation appears in the Results section You will obtain the Predicted value score of the observation described in the table of Explanatory variables as well as the probability that this observation belongs to the target category of the target variable In our example only one variable has been defined The probability that this observation belongs to the target category of the target variable is 0 1120 Note that certain variables of the table of Explanatory variables were automatically completed upon execution of the simulation In fact the model automatically completed certain missing values that were essential to the simulation These values are listing in the following table Type of variable Default value continuous variable the mean value nominal variable the most frequent category ordinal variable the most frequent category These changes are reflected in the left part of the screen after clicking the Run button a SAP InfiniteInsight VX Yy 2 class_Census01 File Help ce Simulating the Model Explanatory Variables Sort by Contribution of class marital status Married civ spouse qe capital gain 592 196 EJ occupation Pros pecialty a education num 9 age 38 6454 a capitalloss 88 5261 rd l l E e ae education HS grad a E
195. hat is the performance of a model compared to a model that would only allow to select observations at random from your database You can thus visualize how much better your model is compared with the random model CUSTOMER SAP Infinitelnsight 7 0 281 2014 SAP AG or an SAP affiliate company All rights reserved Glossary lower bound The term lower bound is defined as an element of P which is lesser than or equal to every element of S M maximum error LInf maximum absolute difference between predicted and actual values upper bound Chebyshev distance mean The arithmetic average value of a collection of numeric data mean absolute error L1 mean of the absolute values of the differences between predictions and actual results City block distance or Manhattan distance mean absolute percentage error MAPE The MAPE value is the average of the sum of the absolute values of the percentage errors It measures the accuracy of the model s forecasts and indicates how much the forecasts differ from the real signal value mean square error L2 square root of the mean of the quadratic errors Euclidian distance or root mean squared error RMSE metadata the information about the data itself CUSTOMER SAP Infinitelnsight 7 0 282 2014 SAP AG or an SAP affiliate company All rights reserved Glossary meta operator Operators that are used upon other operators missing value Data values can be missing beca
196. he Infinitelnsight Scorer feature read the Integration Guide for KMX Generated Codes Caution A specific license is needed to use this feature The code file generated by SAP Infinitelnsight contains all information necessary to the model such as variable encodings missing value replacement compressions model parameters CUSTOMER SAP Infinitelnsight 7 0 184 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification vV To Generate the Code Corresponding to the Model 1 Inthe list Target to be used choose the target of model 2 Use the list nformation to be generated to select the type of results Selected Option Results of the Generated Model Score Estimates score value classification or estimates regression Probability score value and probability value except for HTML and all SQL codes for which only the probability value is provided Bar score value and error bar value except for HTML and all SQL codes for which only the error bar value is provided A Caution Both options Probability and Bar are only available for Infinitelnsight Modeler Regression Classification models with nominal targets l Note In the case of a continuous variable the generated code SQL for example always includes a number of categories that is higher than in the user defined structure or as given by the parameter band count if no user has structure has been set Indeed th
197. he general architecture of SAP Infinitelnsight This section provides an introduction to the elements of this architecture like the various types of interfaces that allow you to use SAP Infinitelnsight BI Dashboard f usthees Oblorte CRM COGNOS Applications Key Drivers rimai Real Forecasting KPI s gt Time SAP Predictive Patterns Analytics panan Data a Warehouse MEOMIAMGR aprimo Mreolane 3 2 1 User Interfaces Three Types of User Interface Three types of interfaces allow you to use the features of SAP Infinitelnsight Graphical user interface Command interpreter API Application Programming Interface controls Graphical User Interfaces The graphical user interface is aimed primarily at end users or non expert users It provides access to the Infinitelnsight modeling assistant which allow you to use SAP Infinitelnsight features and model your data very easily In addition it provides plotted output to facilitate viewing and interpretation of the results of modeling The graphical user interface provided with SAP Infinitelnsight is the Infinitelnsight modeling assistant interface developed in Java on the CORBA API Application Programming Interface which operates on any platform Windows UNIX among others This interface is provided as an example In addition with the Application Programming Interface furnished with SAP Infinitelnsight you can develop your own gr
198. he initial data set are distributed to the estimation sub set 1 4 to the initial data set are distributed to the validation sub set As no test sub set is used all the data from your training data set can be used for sub sets of estimation and validation This can lead to a model with a better quality and robustness Periodic The Periodic cutting strategy is implemented by following this distribution cycle 1 Three lines of the initial data set are distributed to the estimation sub set 2 Oneline is distributed to the validation sub set 3 One line is distributed to the test sub set 4 Distribution begins again at step 1 CUSTOMER SAP Infinitelnsight 7 0 22 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Periodic with Test at the End The Periodic with test at the end strategy distributes 4 5 of the initial data set in a periodic manner to the two sub sets of estimation and validation 3 5 being distributed in the estimation data sub set and 1 5 in the validation data sub set 3 5 being distributed The final 1 5 of the initial data set is sent as a block of data to the test sub set In other words this strategy follows this distribution cycle 1 Three lines of the first 4 5 of the initial data set are distributed to the estimation sub set 2 One line of the first 4 5 of the initial data set is distributed to the validation sub set 3 a lfthe entire 4 5 of the initial data set is not yet
199. he memory store or the folder in which you want to save the model This field allows you to enter the name of the file or table that is to contain the model The name of the file must contain one of the following format extensions txt text file in which the data is separated by tabs or csv text file in which the data is separated by commas SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification setting the Advanced Parameters On the screen Summary of Modeling Parameters click the Advanced button The screen Advanced Model Parameters appears a SAP InfiniteInsight Vx y 7 class_Census01 File Help Advanced Model Parameters General Auto selection Risk Mode Gain Chart Polynomial Degree 1 Score Bins Count 20 Low Prediction Confidence Variables Keep Low Prediction Confidence Variables Exclude Low Prediction Confidence Variables a Correlations Settings Higher than __ 3 ooo 0 50 Keep all Correlations Keep the First 1 024 Target Key Settings Target Target Key class General Tab The General tab allows you to define the general settings of the model that is the degree of the model the score bin count the number of correlations to display and the target key value CUSTOMER SAP Infinitelnsight 7 0 99 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler
200. he records of 1 000 000 prospects identified by their principal characteristics Age Occupation Sex Education Employer Number of hours worked per week Nationality and so on You note that the database you have at hand is not ideal In fact the database contains Incongruous data Redundant data Missing data Incongruent Data The database contains alphanumeric information such as occupation and nationality as well as numerical information such as age and unreconciled accounts Redundant Data Some information in the database is redundant such as degree and education or degree and area of work In the field of statistics the term correlated variables is used to designate such data In classical statistical analyses correlated variables must be processed in a particular manner An alternate solution is to designate only one of the two correlated variables for analysis Since you have neither the statistical skills not the means to handle this issue of correlation between variables you decide to leave the database as it Is CUSTOMER SAP Infinitelnsight 7 0 55 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Missing Data Some information is missing from the database To manage this lack of information the Information Technology department used the following convention The symbol means th
201. he target category 1 of the target variable Class Cluster 2 contains less than 3 3 of customers belonging to the target category In other words Cluster 2 has almost the same customer density belonging to the target category as the data set taken as a whole Cluster 8 is the cluster with the lowest density of observations belonging to the target category Compared to the entire data set Cluster S contains 23 5 fewer of customers belonging to the target category This cluster therefore has a density of customers belonging to the target category lower than that of the data set Clusters Profiles Cross Statistics and Variables Profiles The clusters profiles allow you to view for each cluster the profile of each explanatory variable with respect to their profile over the entire data set the SQL expression of the cluster when they have been calculated Variable Profile The Variable Profile indicates the distribution of observations belonging to a cluster of global data set within the categories of each variable In other words the profile indicates the proportion of observations contained in each of the categories of that variable Example of a Variable Profile The variable gender of a data set can be distributed as follows 53 of observations belong to the category male 47 of observations belong to the category female This distribution corresponds to the profile of the variable gender over this data set Gi
202. his profile To implement this method the statistician must Perform a detailed analysis of your test campaign Prepare your database down to the smallest detail specifically encoding the different types of data in such a way that they can be used by the analytical tools he will apply Test different types of algorithms for example neural networks Bayesian networks logistic models decision trees and select the one best suited to your business Issue Typically after a few weeks the statistician will be able to associate a value with each individual in your database indicating the probability of being interested or not interested in your marketing campaign This method presents significant constraints You must Ensure that your statistical expert perhaps from a department external to the Marketing Department is available for the scheduled period Ensure that the cost for using this scarce resource will fit into your budget Spend time explaining your domain specific business issue to him Spend time understanding the results that are provided Infinitelnsight Method The simplicity and automatic nature of SAP Infinitelnsight will allow you to perform the statistical analysis of your database yourself In addition using SAP Infinitelnsight will allow you to obtain results in mere minutes SAP Infinitelnsight uses the latest innovations of statistical sciences and also liberates the end user from the
203. i Browse Data r W Browse Define Mapping Ed The input file is missing 41 Previous I gt Apply 3 Click the Browse button to select 174 In the Folder field the folder which contains your data set In the Data field the name of the file corresponding to your data set CUSTOMER 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification SAP Infinitelnsight 7 0 4 Inthe section Results generated by the model select the file format for the output file l Note The current version of SAP Infinitelnsight does not allow you to save the file in an ODBC database Click the Browse button to select In the Folder field the folder in which you want to save the output file In the Data field the name that you want to give the output file In the Generate Field select the type of output values that you want for the target variable 5 You may also opt to select Save only outlier observations If you select this option only the outlier observations will be presented in the results file obtained after applying a model 6 Click the Apply button The screen Applying the Model appears Once application of the model has been completed the results files of the application is automatically saved in the location that you had defined from the screen Applying the Model E DAP InfiniteInsight Vx y 7 class_Census01 File Help Applying the Model
204. iable Cluster 6 vs Whole Population for Variable capital gain 0 75 ai Sete 29 0 50 Significance KL 1 0 25 Overall mean 592 196 mea Cluster mean 10163 4 v q ah ah a i a wa p gt A Categories E I Population Cluster 6 44 Cance E IMR CUSTOMER SAP Infinitelnsight 7 0 256 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering In the figure above the table allows you to identify cluster 6 as the cluster containing the highest density of observations belonging to the target category of the target variable 88 8 of customers contained in this cluster belong to target category 1 of the target variable Class The cross statistics plot allows you to view and compare the profiles of the variable capital gain over the entire data set and over cluster 6 These profiles are repeated in the table below Categories Profile over the Profile over of the variable capital gain data set cluster 6 KxMissing 1 6 O 92 0 JO 4386 3 0 J4386 41310 5 91 The data distribution of the category 4650 41310 makes it clear that the majority of customers contained in cluster 7 realize significant annual capital gains relative to the entire set of customers contained in the data set In addition the data distribution over the category O indicates that the majority of the customers contained in the data set or 92 do not realize any annual capital g
205. iables Click the link indicating the number of variables in the sentence Each step removes1 variable A slide is displayed ranging from 1 to the total number of variables in the model Move the cursor on the slide to select the number of your choice Click OK To Select the Information Amount Click the link indicating the amount of information to keep in the sentence Lach step keeps 95 0 of information A slide is displayed Move the cursor on the slide to select the quantity of your choice Click OK To Set the Authorized Quality Loss The quality loss can be set in the sentence Search process stops with a drop of 1 0 ofKI and KR A N a 106 Click the link indicating the percentage of loss for example 5 0 A slide is displayed Select the maximum percentage of authorized quality loss with the cursor Click OK Click the quality criterion A drop down list is displayed offering the following options Based on KI 2 KR the quality loss is based on both the predictive power KI and twice the prediction confidence KR Kland the KR the quality loss is limited for both the predictive power KI and the prediction confidence KR It is the default value KI the quality loss is limited for the predictive power KI only KR the quality loss is limited for the prediction confidence KR only Select the option of your choice Click OK CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP
206. iables For this Scenario Select the variable marital status which is the explanatory variable that contributes the most to the target variable Class a SAP InfiniteInsight VX y 2 cass_Census01 File Help ally Category Significance A E 9 Ams A Data Sets View Type a Copy Print Save EXPOrtto pin view ae tet eee Orientation r Excel Variables marital status w Variable marital status Influence an Target 0 3 0 2 0 1 0 0 0 1 0 2 0 3 0 4 0 5 Married AF spouse Married civ spouse Divorced Maried spouse absent Categories Separated Widowed Never married This plot presents the effect of the categories of the marita status variable on the target variable CUSTOMER SAP Infinitelnsight 7 0 132 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Variable Categories and Profit The plot Category Significance illustrates the relative significance of the different categories of a given variable with respect to the target variable On this type of plot The higher on the screen one finds a category the greater the positive effect on the target category or hoped for value of the target variable In other words the higher a category appears on the screen the more representative that category is of the target category of the target variable The width and direction of the bar correspond to the profit contr
207. ibbon 3 Click Edit User Band Count fe SAP InfiniteInsight Vx y z New Regression Classification Model Iof x File Help Data Description Main Edition Structures a New Structure Edit User Band Count From Statistics From Model r Remove Structure From Variable E Description Desc_Census01 csv Sr TG Bfnlwot_ number continuous 0 oo o o l oOo a Sleducation n number ordinal 0 4 Ascreen is displayed as below W Set Band Count 2s Count hours per week 0 hours per week 0 Set the Same Band Count for All Variables 20 CUSTOMER SAP Infinitelnsight 7 0 87 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification If you want to Then modify the band count for all the 1 Type in the desired band count in the field at the bottom of the panel continuous variables of the model 2 Click Set the Same Band Count for All Variables 3 Click OK modify the band count for the variable 14 Type in the desired band count in the column Band Count at the top of the being edited panel 2 Click OK Optimal Grouping for All Variables When working with a defined structure if want to keep your categories as they are defined for the model building you must disable this option If not or if you work with no defined structure Optimal Grouping allows in a large number of cases to increase the prediction confidence of the model with a minimal l
208. ibuted by that category In other words they correspond to the relationship of that category to the target variable and whether that category has more or less observations belonging to the target category of the target variable For a given category a positive bar on the right of 0 0 indicates that the category contains more observations belonging to the target category of the target variable than the mean calculated on the entire data set A negative bar on the left of O O indicates that the category contains a lower concentration of target category of the target variable than the mean l Note You can display the profit curve for the selected variable by clicking the button we Display Profit Curve located in the tool bar under the title The importance of a category depends on both its difference to the target category mean and the number of represented cases High importance can result from a high discrepancy between the category and the mean of the target category of the target variable or a minor discrepancy combined with a large number of records in the category or acombination of both The width of the bar shows the profit from that category The positive bars correspond to categories which have more than the mean number from the target category that is responders and the negative bars correspond to categories which have less than the mean number from the target category that is responders The Variables pull down me
209. icity Gender Annual capital gains Annual capital losses Country of origin Variable indicating whether or not the Salary of the individual is greater or less than 50 000 Example of Values Any numerical value greater than 17 Private Self employed not inc Any numerical value such as 0 2341 or 205019 lith Bachelors A numerical value between 1 and 16 Divorced Never married Sales Handlers cleaners Husband Wife White Black Male Female Any numerical value Any numerical value United States France 1 if the individual has a salary of greater than 50 000 0 if the individual has a salary of less than 50 000 In order to avoid complicating the Infinitelnsight Modeler Regression Classification and Infinitelnsight Modeler Segmentation Clustering application scenarios the variable fn wgt is used as aregular explanatory variable in these scenarios and not as a weight variable CUSTOMER 202 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering SAP Infinitelnsight 7 0 6 1 7 Infinitelnsight modeling assistant To accomplish the scenario you will use the Java based graphical interface of SAP Infinitelnsight This interface allows you to select the SAP Infinitelnsight feature with which you will work and help you at all stages of the modeling process vV To Start Infinitelnsight modeling assistant 1 Select Sta
210. idation The default graphic displays the actual target values as a function of predicted target values Two curves are displayed one for the Validation sub set blue line and another for the hypothetical perfect model Wizard green line The Validation curve gives Actual Target value as a function of Predicted Target value For example when the model predicts 35 the average actual value is 37 The Wizard curve Is just X Y meaning that all the predicted values are equal to the actual values The graph is an easy way to quickly see model error When the curve is going far from Wizard it means that the predicted value is suspicious The graph is computed as follow about 20 segments or bins of predicted values are built Each of these segments represents roughly 5 of the population for each of these segments some basic statistics are computed on actual value such as the mean of the segment SegmentMean the mean of the associated target TargetMean and the variance of this target within that segment TargetVariance For example for predicted value in 17 19 the mean would be 18 5 the actual target mean would be 20 5 and the actual target variance would be 9 In this case we could say that if the predicted value is between 17 and 19 the model is underestimating a bit the actual value For each curve a dot on the graph corresponds to the segment mean on the X axis and the target mean on the Y axis The blue area represen
211. ies from the variable statistics then editing or validating the suggested structure by importing the structure from an existing model by building a new structure from scratch The option Optimal Grouping allows you to let Data Encoding group together the categories groups defined in the variable structure if they bring the same information The last column of the description table indicates the state of the structure of each variable The following table lists the possible states of a variable structure Icon State Description r undefined Data Encoding will automatically determine the categories grouping depending on their interaction with the target variable g non editable The structure for an ordinal string variable cannot be modified a defined by extraction from the variable statistics The user must open and validate the variable structure defined by the user or imported from an existing model l Note A translation of the variable categories has no influence on the variable structure which has to be set according to the original values of the variable CUSTOMER SAP Infinitelnsight 7 0 76 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification vV To Extract a Variable Structure 1 Select the variables for which you want to extract the structure a SAP InfiniteInsight VX y 2 New Regression Classification Model File Help Data Description Ma
212. ilter to the data set you can use a file created during a previous use of the data setin SAP Infinitelnsight 1 Click the button Load Existing Filter A pop up window is displayed 2 Use the list Data Type to select the format of the filter 3 Use the Browse button located on the right of the Folder field to select the folder or the database in which the filter is stored 4 Use the Browse button located on the right of the Description field to select the file or the table containing the filter 5 Click OK Translating the Variable Categories You can translate the categories of a nominal variable save the translation or load an existing translation This translation has no influence on the variable structure which has to be set according to the original values of the variable l Note The variable Target Key that is used in the advanced settings for example does not take into account the translation when displaying the possible values of this variable vV To Translate the Variable Categories 1 Click a nominal variable to translate its categories 2 Goto the Edition tab of the ribbon and click the option Translate Categories A new window appears 3 Choose into which languages you want to translate By default the language of the user interface is displayed as a column 4 Clickthe button to extract the variable categories from the data set 5 Translate the categories l Note You do not need to fill all fields
213. in Edition Structures A l Open Description Save in Variable Pool Q i Analyze kH Save Description Remove from Variable Pool View Data Properties Description i Description Desc_Census01 csv Index Name Storage Value Key Order Missing Group Description Structu re jjage number continuous 0 0 a 2 workclass string nominal 0 0 7 a 3 fniwat number continuous 0 0 a leducation sting nominal 0 5 education n number ordinal 0 0 a 6 marital status string nominal 0 0 Z T occupation string nominal 0 0 a relationship string nominal 0 0 a grace string naminal 0 0 a 10 sex string nominal 0 0 a 11 capital gain number continuous 0 0 99999 a 12 capitalloss number continuous 0 0 a 43 hours per w number continuous 0 0 a 14 native country string nominal 0 0 a 15 class number nominal 0 0 F 16 Kxlndex integer continuous 1 0 Automaticall aa Add Filter in Data Set keo oo CUSTOMER SAP Infinitelnsight 7 0 a 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 2 Goto the Structures tab of the ribbon the available options are separated in two parts Edit and Extract fe SAP InfiniteInsight Vx y File Help 7 New Regression Classification Model Data Description dh New Structure Edit User Band Count From Statistics From Model r Remove Structure From Variable ee tage number
214. in each cluster The figure below presents the Target Means plot obtained in during this scenario The bars have been sorted in descending order Target Means Data Set Estimation 6 0 85519775695801 Clusters E Target Means Among the five clusters C uster 6 is the one that has the greatest proportion of observations belonging to the target category of the target variable In fact 85 5 of the customers contained in cluster 6 belong to target category 1 target variable Class In other words 85 5 of the customers contained in cluster 6 responded in a positive manner to the test phase of your marketing campaign Cluster amp is the cluster with the lowest density of observations belonging to the target category Less than 1 of the customers contained in this cluster responded positively to the test phase of your marketing campaign CUSTOMER SAP Infinitelnsight 7 0 250 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Tne Plot Frequencies The Frequencies plot presents the number of observations contained in each cluster relative to the total number of observations contained in the data set The figure below presents the Frequencies plot obtained during this scenario The bars have been sorted in descending order Frequencies Data Set Estimation 0 250 0 225 0 200 O 175 0 150 0 125 0 100 O 075 0 050 0 025 0 000 Clusters E Frequenci
215. ined At this step the statistician may unconsciously bias the results Test different types of algorithms K means both ascending and descending hierarchical segmentation models and select the one best suited to your business Issue Evaluate the relevance of the clusters obtained in particular the response to your domain specific business issue After a few weeks the statistical expert will be able to provide a certain number of clusters or homogeneous groups to which each of the individuals of your database are assigned This method presents significant constraints You must Ensure that your statistical expert who is usually from an external department is available for the scheduled period Ensure that the modeling costs will fit into your budget Spend time explaining your domain specific business issue to the statistician Spend time understanding the results that are provided Aska programmer to write a program to determine the cluster associated with any new individual added to your database In addition this method is not systematic Two statisticians performing this segmentation on the same data set could obtain different results Infinitelnsight Method Infinitelnsight Modeler Segmentation Clustering allows you to build a segmentation model of your customers in a few minutes taking into consideration the interest expressed by your customers in your new product Infinitelnsight Modeler Se
216. ing data set in order to Rapidly create an explanatory and predictive model Next apply this model to the entire database Using the model generated you will be able to determine How many individuals contained in your prospects database you should send your mailing to in order to maximize the profit return on investment of your campaign How to classify all of the individuals in your prospects database according to their interest purchasing probability in this new product This interest is expressed as a score or probability that a prospect will respond favorably to the campaign What characterizes these individuals and what is their profile Validate the criteria age socio occupational class degree that explain why a person expresses interest or not in the new financial product How to simulate in real time the likelihood of a single individual to respond favorably to a new offer in particular to allow the Call Center of your bank or a customer service agent to immediately know the level of interest that a prospect is likely to exhibit in this financial product How to record this Score in your prospects database in order to be able to select sub groups of the population for new campaigns at a later date _ How to measure the quality and reliability capacity of handling new individuals of your model In order to allow you to better respond to these issues you have access to several possible
217. ing strategy determines the way in which the data of the training data set are distributed across the sub sets The Estimation and Validation sets are used for actual training and the Test set Sometimes referred to as the hold out sample is used to ensure that the predicted performance is correct l Note When using SAP Infinitelnsight the data sub sets are virtual they are not stored in memory at any time The file corresponding to the initial data set remains intact at all times The figure below illustrates the model generation process known as the training phase Validation sub set Training f data set BEA a ae a Test sub set 4 7 5Representation of a Model A model may be represented in many different ways including a decision tree aneural network amathematical function In SAP Infinitelnsight models are represented in the form of mathematical functions specifically polynomials Description of the Polynomial A polynomial may be of degree 1 2 3 or greater By defining the polynomial degree you are defining the degree of complexity of the model CUSTOMER SAP Infinitelnsight 7 0 36 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Examples of Polynomials A polynomial of degree 1 is of the form f X1 X2 bey XN w0 w1 X1 w2 X2 wWn Xn A polynomial of degree 2 is of the form f X1 X2 Xn w0 wl1 X1
218. ing the Model Model Name class_Census01 Description Data Type Tex Files Folder Samples Census File Table 44 Cancel FJ You must select a fileftable 41 Previous Ib Save 2 Complete the following fields Model Name This field allows you to associate a name with the model This name will then appear in the list of models to be offered when you open an existing model Description This field allows you to enter the information of your choosing such as the name of the training data set used the polynomial degree or the predictive power and prediction confidence obtained for the model This information could be useful to you later for identifying your model Data Type this list allows you to select the type of storage in which you want to save your model The following options are available Text files to save the model in a text file Database to save the model in a database Flat Memory to save the model in the active memory SAS Files to save the model in a SAS compatible file for a specified version of SAS and a specified platform SAS v6 or 7 8 for Windows or UNIX SAS Transport to save the model in a generic SAS compatible file Folder Depending upon which option you selected this field allows you to specify the ODBC source the memory store or the folder in which you want to save the model File Table This field allows you to enter the name of the file or table th
219. initeInsight class_census_apply2 all Statistical Reports HE asua s Kar ngine e J MyConstant Real_estate Sport car H pool _ E Grouped Cross Statistics with the T H Model Performance H Control for Deviations amp Expert Debriefing E Groups Id SESUISeIE TS H Model Settings CUSTOMER SAP Infinitelnsight 7 0 138 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Target Specific Exclusions showing the variables excluded towards a particular target A SAP InfiniteInsight Vx y Z fnlwgt_Census01 File Help al Statistical Reports E a aS S View Sort Series Copy Print Export bl ho al T T T F ES Target fi ee E Variables Variable Reason for Exclusion C_capital gain Small Kl On Estimation c_capitaltoss Small Kl On Estimation capital gain Small Kl On Estimation capital loss Small KI On Estimation class Fully Compressed 2 El Category Frequencies gt E Category Significance E Continuous Variables pen im Continuous Targets Number l Data Set Size H E Variables Correlations ae Cross Statistics with the Targeti H Grou ped Cross Statistics with tl H Model Performance He Control for Deviations E El Expert Debriefing o z l Groups Id E Other Variables Performance In fin Continuous Encoding EP Variables Exclusion Cause Se arget Specific Exclusions H Mo
220. ion The Significance of Categories plot illustrates the relative significance of the different categories of a given variable with respect to the target variable Displaying the Significance of Categories Plot vV To Display the Significance of Categories Plot 1 2 240 On the screen Using the Model click Category Significance The plot Category Significance appears a SAP InfiniteInsight Vx y z class_Census01 File Help aly Category Significance A wd 9 Asb a j Export to Copy Print Save padid h Excel Pin View Variables age w Variable age Influence on Target 0 075 0 050 0 025 0 000 0 025 0 050 0 075 0 100 0 125 0 150 0 175 0 200 categories Ta ha a 19 22 Hi EE a a E Validation In the Variables list located above the plot select the variable for which you want to display the categories If your data set contains date or datetime variables automatically generated variables can appear in the Variables list For more information refer to section Date and Datetime Variables Automatically Generated Variables on page 30 CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering l Note You can display the relative significance of the categories of a variable directly from the plot Contributions by Variables On the plot Contributions by Variables do
221. ion Operation of SAP Infinitelnsight Overview which provides a general description of the use of SAP Infinitelnsight in the model generation process IN THIS CHAPTER Operation of SAP Infinitelnsight OVeErVview cccccccceccccccceesseeseeceeeeeeeeeeseeeeeeessaeeueeeeeeesseaeaeeeeeesssaeaaeeeeeeessaaaeees 17 Data SOUMCCS SUD DONC sssr ced acetcidirccjen cate sdicci neat et E EE N aE lle EE o NEO xe o ANE 18 A hs SO iach cate get E E cena A ace nd tee ase lea eens in ence a eaten gutted one canter dea E E E E eae 18 Gne kE eS CS sacs aca E E E E ace E ence dae E E E E E A E E 19 AO OT DA ea E E E tie enesenianetarieaae tone 24 W ANAS 6 ar E E E E E E 25 MoO ee ec ee a Pe ee ee ee eee na een tec ne ee ree 35 Pernoimance ele ere 0 S sossa E eam eT Sk eee ve aT Sree anny ety eer een ee ene eee eee a 39 BU MM DO oe steppe cote A Soe oeee eaten aba nee A A caeseoes vara ateanieneoscaneesed baa eeeasoeaede 46 Advanced MOG SOI SS seeneniit enaa a aE E T AE Ea E eaa Aani 47 CUSTOMER SAP Infinitelnsight 7 0 16 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 1 Operation of SAP Infinitelnsight Overview SAP Infinitelnsight allows you to perform supervised Data Mining that is to transform your data into knowledge then into action as a function of a domain specific business issue SAP Infinitelnsight supports various formats of source data flat files ODBC compatible sources In order to b
222. ion structure 4 Click View Type and select the 2d Profiles button to go back to the Cross Statistics plot CUSTOMER SAP Infinitelnsight 7 0 258 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Understanding SQL Expressions The SQL Expressions screen can be broken down into two parts n the upper part a table presents each cluster in a Summarized fashion It allows you to select the cluster for which you want to display a SQL expression Inthe lower part a tree presents the SQL expression corresponding to the selected cluster The following schema presents the SQL expression for Cluster 3 a SAP InfiniteInsight Vx y z cass_Census01 File Help ali Cluster Profiles a E Ade E View Type Bar Sort Reset Cluster Copy Print Save Export to Orientation Categories Names Excel Cluster Name Frequencies of 1 1 f 4 22 ee 51 17 2 RE 25 24 D 27 16 4 244 E 2 23 5 ie 67 57 8 BB 5 5 ee 25 51 7 es 4 ao E E E Variable ranges for Cluster 3 Fl AND Z marital status in Never married Married spouse absent Divorced Separated Widowed 1 bn age in J24 27 28 30 31 32 33 34 35 36 37 38 39 40 41 43 44 45 46 90 occupation in Handlers cleaners Adm clerical Priv house sery Armed Forces Other service Sales Craft repair Transport moving Farr
223. ipt Only one file is generated Save the Description with the Script the data description is saved in an additional file in the same folder as the KxShell script Save the Description with the Data the data description is saved in an additional file in the same folder as data used for the model Save the Description Separately the data description is saved in an additional file The user indicates the type of the description text file data base flat memory and the location where the data description should be saved l Note When saving the description in an additional file the file is named following this syntax KxDesc_ lt Dataset Role gt _ lt Dataset Name gt For example for a training data set named Census csv the description file name will be KxDesc_Training_Census csv CUSTOMER SAP Infinitelnsight 7 0 190 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Additionally you can export the variable structure with relation to a target variable by checking the option Export Variable Structure in Script and selecting the target variable in the list Select a Target This option allows you to force the grouping of categories when training the model on new data sets Before exporting the script you can view the script by clicking the button Script Preview oye B Sn a en ee a ee a tchangeParameter Parameters VariableSelection StopCriteria MaxNbOfFinalVariables
224. is Document 2 Welcome to this Guide IN THIS CHAPTER About this DOCUMEN cette gece terete aos etree a Oa Eana a cts cua Ea sew ncee ute aleerystewsinneuconceeu ies BOOS BENNING pemeemereme emma ie ree etre aner sere ter nCee ener eee eve ee eer tate re mein E rave ee heey eee eer 2 1 About this Document 2 1 1 Who Should Read this Document This document is addressed to people who want to evaluate or use SAP Infinitelnsight 2 1 2 Prerequisites for Use of this Document Use of this guide does not require any prior expertise in statistics or databases SAP Infinitelnsight features are developed using cutting edge technologies and while they call on complex innovative statistical techniques they are still straightforward and quick to use they put powerful Data Mining techniques within the reach of non expert users For more technical details regarding SAP Infinitelnsight please contact us We will be happy to provide you with more technical information and documentation CUSTOMER SAP Infinitelnsight 7 0 7 2014 SAP AG or an SAP affiliate company All rights reserved Welcome to this Guide 2 1 3 What this Document covers This document introduces you to the basic concepts underpinning SAP Infinitelnsight and the main functionalities of the Infinitelnsight Modeler Regression Classification and Infinitelnsight Modeler Segmentation Clustering features Using two application scenarios you can create your first mo
225. is declared as date or datetime the date coder feature automatically extracts date information from this variable such as the day of the month the year the quarter and so on Additionnal variables containing this information are created during the model generation and are used as input variables for the model The date coder feature is disabled for Time Series Type continuous nominal ordinal or textual For more information about data description see Types of Variables on page 26 and Storage Formats on page 29 How to Describe Selected Variables To describe your data you can Either use an existing description file that is taken from your information system or saved from a previous use of SAP Infinitelnsight features Or create a description file using the Analyze option available to you in Infinitelnsight modeling assistant In this case it is important that you validate the description file obtained You can save this file for later re use If you name the description file KxDoc_ lt SourceFileName gt it will be automatically loaded when clicking the Analyze button A Caution The description file obtained using the Analyze option results from the analysis of the first 100 lines of the initial data file In order to avoid all bias we encourage you to mix up your data set before performing this analysis CUSTOMER SAP Infinitelnsight 7 0 72 2014 SAP AG or an SAP affiliate company All rights reserved
226. is divided into several parts with each part in turn used to test a model fitted to the remaining parts customized cutting strategy The customized cutting strategy allows you to define your own data sub sets To use this strategy you must have prepared before opening SAP Infinitelnsight features three sub sets the estimation validation and test sub sets customized profit Customized profit allows you to define your own profit values that is to associate both a cost and a benefit to each value of the target variable CUSTOMER SAP Infinitelnsight 7 0 275 2014 SAP AG or an SAP affiliate company All rights reserved Glossary cutting strategy A cutting strategy is a technique that allows decomposition of a training data set into two or three distinct sub sets An estimation sub set A validation sub set Atest sub set This cutting allows for cross validation of the models generated D data aggregation The process of consolidating data values into a smaller number of values For example sales data could be collected on a daily basis and then be totaled to the week level data set a collection of data usually presented in tabular form where each column represents a particular variable and where each row is an assignment of values data source A data source includes both the source of data itself such as relational database a flat file database or evena text file and the connection information n
227. ision 68 28 True 0 94868 ie ee rN i o oo 68 65 PESENE nee Total Po pulation 12 461 Cost Matrix Predicted 1 Predicted 0 gt Profit 0 True 4 Random 0 Maximize Profit True 0 Gain 0 Definitions A positive observation is an observation that belongs to the target population A negative observation is an observation that does not belong to the target population CUSTOMER SAP Infinitelnsight 7 0 144 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Understanding the Confusion Matrix There are three ways to set the threshold using the displayed slide bar by selecting the percentage of population to target if the population is sorted by descending order of score of Population by selecting the percentage of positive observations you want to detect of Detected Target by selecting the score used to differentiate positive observations from negative ones Score Threshold Any observation with a score above the threshold Is considered positive on the contrary any observation with a score below the threshold is considered negative The slide is graduated from the lowest score on the left to the highest score on the right The values corresponding to each option are displayed is displayed under the slide When you move the cursor the confusion matrix is updated accordingly The following table details how to read the confusion matrix
228. iterion gt _ lt threshold gt _ lt rank gt _rr_ lt target name gt contains the value of the reason code that is the difference between the variable contribution for the customer and the threshold CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Continuous larget Option Output Column Name This option allows you to Predicted Value rr_ lt target variable gt generate in the output file the value predicted by the model for the target variable This option is checked by default Confidence bar_rr_ lt target variable gt add to the output file the confidence level for the value that has been predicted this is also known as the error bar Itis computed with 3 standard deviations on the validation data set and bin per bin The percentage of population corresponding to the 3 standard deviations is of about 99 Calculation formula TargetMean 3 sqrt TargetVariance TargetMean 3 sqrt TargetVariance When sqrt TargetVariance is equal to the Standard Deviation TargetMean Standard Deviation is equal to the Confidence Interval Outlier outlier_rr_ lt target variable gt to show in the output file which observations are outliers An Indicator observation is considered an outlier if the difference between its predicted value and its real value exceeds the value of the error bar In other words the error b
229. k the day of week according to the ISO disposition Monday 0 lt OriginalVariableName gt _DoW and Sunday 6 Day of month the day of month 1 to 31 lt OriginalVariableName gt _DoM Day of year the day of the current year 1 to 366 lt OriginalVariableName gt _DoY Month of quarter the month of the quarter January April July and lt OriginalVariableName gt _MoQ October 1 February May August and November 2 March June September and December 3 Month of year the month 1 to 12 lt OriginalVariableName gt _M Year the year lt OriginalVariableName gt _Y Quarter the quarter of the year J anuary to March 1 April to lt OriginalVariableName gt _Q June 2 July to September 3 October to December 4 From datetime variables Temporal Information Represents Generated Variable Name Hour the hour lt OriginalVariableName gt _H Minute the minute lt OriginalVariableName gt _Mi Second the second lt OriginalVariableName gt _S H seconds the micro second lt OriginalVariableName gt _mu The generated variables will appear in the model debriefing panels listing variables such as the Contributions by Variable the Category Significance the Statistical Reports as well as in the automatic variable selection feature 4 6 5 Roles of Variables In data modeling variables may have three roles They may be Target variables Explanatory variables Weight variables CUSTOMER SAP Infinitelnsight 7 0 3
230. kely to be interested in the new financial product Identifying the ideal number of prospects to contact out of the entire database Using the Infinitelnsight Modeler Regression Classification formerly known as K2R feature you can rapidly develop an explanatory and predictive model at the least possible cost This model allows you to respond to your business issue and accomplish your objectives 5 1 2 Your Objective Imagine the following case You are the Marketing Director of a large retail bank This bank has decided to offer its customers a new high end savings product It prepares to launch an extensive direct marketing campaign to promote this new product to its prospects and customers The bank is experiencing heavy competition and senior management sensitive to the stakes involved in launching this new financial product wants the marketing campaign completed as soon as possible CUSTOMER SAP Infinitelnsight 7 0 54 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 5 1 3 Your Means A Limited and Closely Monitored Budget The enterprise controls of the bank are rigorous and the budget that has been allocated to you for this marketing campaign Does not allow you to contact all of the bank s prospects and customers May not be exceeded The Information at your Disposal The Marketing Department has a database for this campaign which contains t
231. l is a critically important phase in the overall process of Data Mining Always be sure to assign significant importance to the values obtained for the predictive power and the prediction confidence of a model CUSTOMER SAP Infinitelnsight 7 0 229 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering To validate a segmentation model you can also observe the value of the indicators frequency and target mean for each of the identified clusters Specifically the most interesting clusters of the segmentation model will possess an elevated frequency anda target mean that deviates from the target mean of the entire data set Note that a segmentation model with a low predictive power may conceal precisely this type of cluster To find out how the frequency and target mean for a cluster are calculated see Understanding the Detailed Description of Clusters For this Scenario The model generated possesses A predictive power equal to 0 703 A prediction confidence equal to 0 987 The model performs sufficiently well You do not need to generate another To Validate the Model Generated 1 Verify the Predictive Power KI and the Prediction Confidence KR of the model These indicators are highlighted on the following figure a SAP InfiniteInsight Vx y z class_Census01 File Help Training the Model a sa Stop ViewType Copy Print Save Export to T Power
232. l rights reserved Infinitelnsight Modeler Regression Classification The Model Name is filled automatically It corresponds to the name of the target variable class for this scenario followed by the underscore sign _ and the name of the data source minus its file extension Census01 for this scenario You have the possibility to display the results generated by K2R as a decision tree based on the five most contributive variables To activate this option check the box Compute Decision Tree The Autosave button allows you to activate the feature that will automatically save the model once it has been generated When the autosave option is activated a green check mark Is displayed on the Autosave button Activating the Autosave Option The panel Model Autosave allows you to activate the option that will automatically save the model at the end of the generation process and to set the parameters needed when saving the model vV To Activate the Autosave Option 1 Inthe panel Summary of Modeling Parameters click the Autosave button Compute Decision Tree Enable Auto selection 4 Previous The panel Model Autosave is displayed 2 Check the option Enable Model Autosave a SAP InfiniteInsight Vx y 7 class_Census01 File Help Model Autosave a a a a a a a ma aa m a a a S D A s a m m Description Data Type Tex Files wt Folder J SamplesiCensus 44 Cancel EJ You must select a filef
233. lated Variables To say that variables are correlated implies a certain level of redundancy that they each contribute some of the same information with respect to the target variable Two variables said to be highly correlated would describe the same information or the same concept to an even greater degree The plot Smart Variable Contributions reflects the correlation that may exist between various explanatory variables When two variables A and B are strongly correlated Variable A with a greater contribution than B with respect to the target variable becomes the primary variable the plot displays all its information including what it has in common with variable B Variable B with a smaller contribution than A with respect to the target variable becomes the secondary variable only its marginal contribution is displayed on the plot meaning that only the supplementary contribution to target variable information or the values that B does not share with A are displayed This difference of information is noted variable_B variable_A Encoded Variables Creating an SAP Infinitelnsight model uses not only the original variables but also in case of continuous or ordinal variables their value as encoded by Infinitelnsight Modeler Data Encoding This is called dual encoding and allows SAP Infinitelnsight to find all the information contained in each variable The encoded variables appear on the variable contributions pl
234. le A probability over 0 95 indicates that there is a change of behavior with respect to the target variable in the category or group of categories the last group contains only the option Category with Problem For each data set reference data sets and control data set all variable categories with a probability over 0 95 are listed This allows you a quick visualization of possible problems without having to analyze all the reports A Caution In all the report panels the control data set is referred as the Apply n data set i SAP InfiniteInsight VX y 2 cass_Census01 File Help Control on Application Data Set jm Jya A amp 8 View Sort Seres COPYCurment Print Export T IEW H Detailed Statistics on Control Grou Empty Report for Deviant Variables Cc On Options You can select which report sections to save 1 Click the button k Save the reports located in the bottom left corner A selector window opens Select the Report Sections to be Saved Ea v amp Control on Application Data Set E v amp Summary vV i Deviant Variables E v amp Deviation Detailed Statistics je N E Probability of Deviation on Application Data Set v B Probability of Category Deviation on Application Data Set V l Probability of Target Deviation on Application Data Set V l Probability of Grouped Category Deviation on Application Data Set El v amp Detailed Statistics on Control Group
235. le salary in US dollars 1000 00 1593 and 2000 54 The variable age in years 21 34 and 99 The variable family name Lake Martin and Miller The variable occupation professor engineer and translator The variable telephone 800 555 1234 and 800 555 4321 A variable that has numbers for values is not forced to be described using the number storage format For instance the variables telephone and zip code may instead be described using the string storage format because no arithmetic operations that make any sense can be performed on these values Similarly a variable that will be used as an observation identification code in a table and does not comply with supported number formats may be described using the string storage format Caution For number storage formats the decimal separator used must be a decimal point and not a comma So the value 6 5 may be processed while 6 5 will not be processed CUSTOMER 29 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight 7 0 Essential Concepts Date and Datetime Variables Automatically Generated Variables When your data set contains date or datetime variables the feature Date Coder automatically extracts date information The Date Coder is able to extract the following temporal information For date or datetime variables Temporal Represents Generated Variable Name Information Day of wee
236. le Help Data Description iB Open Description Save in Variable Pool a i Analyze Save Description Remove from Variable Pool View Data Properties Description Index Name Storage Value Key order Missing Group TDescription Siruchi re _ Add Filter in Data Set 44 Cancel Eg You must analyze the data or open a description file 41 Previous I Next 3 Goto section Describing the Data Selected CUSTOMER SAP Infinitelnsight 7 0 68 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Special Case Data Stored in Databases the Explain Mode Before requesting data stored in a Teradata Oracle or SQLServer 2005 database SAP Infinitelnsight uses a feature called the Explain mode which categorizes the performances of SQL queries in several classes defined by the user In order to be as fast and as light as possible this categorization is done without actually executing the full SQL query l Note 1 For all versions of Teradata 2 For all versions above and including Oracle 10 The objective is to allow estimating the workload of the SQL query before executing it and then deciding possibly thanks to an IT Corporate Policy if the SQL query can actually be used For example an IT Corporate Policy may favor interactivity and then define 3 classes of SQL queries each with its maximum time mmediate duration lt
237. ler Regression Classification ccsssssssccssnssseceeensseeeeeenssseceennsssseeeenssseees 54 Application Scenario Enhance Efficiency and Master your Budget using Modeling ccceeeeeeesseeeeeeeees 54 5 1 1 PUCSCHILAU OM sesanan a NEE ONON 54 SLZ TOUF ODJECUVE nAn AaB A E Pn OE ee oe 54 5 1 3 OUP IVICGINS aira e ete cere lene er ine eee ee er Le ee hae enaneaecenss 55 5 1 4 OUI PAD OF OAC heee ne nee EEE E E E suevores sewees ueeueces AAS 56 5 1 5 OUD USINCSS Io UE heee A a inst atte hee tebe 57 5 1 6 YOU SOIUHONS etme on E A A AE PP 57 Oley mtrod ctiontosample File Serrana e teaches Meena ante Atlas Aaa E al ea ek a ol 60 5 1 8 Infinitelnsight modeling assistant sivzesessintasetsbetvandstetoned dncorictcente te ceandttseecgsaeettty detobtaceteceeeatceezett 62 Creating a Classification Model Using Infinitelnsight Modeler 0 ccccccecsssssceeeeeecececeeeeeeeeeetsssssseeeeeeeees 66 Diz Step 1 Defining the Modeling Par annette rs cit tte cscs ee 66 5 2 2 Step 2 Generating and Validating the Model eccrie te tel et ei eth ha Bl ee Bat 109 5 2 3 Step 3 Analyzing and Understanding the Model Generated ccc cceeecessssssssssssssssseeeeeeeeeeeeeeeeens 113 5 2 4 SECA UW SHS MEMOGA E NAANA ANR 154 Infinitelnsight Modeler Segmentation Clustering cccsssssseccssnsssececenssseeeeenssseeeeensssseesenssseees 197 Application Scenario Customize your Communications using Data MOdeling cc cc
238. lication copies the parameters of the plot You can paste it into a spreadsheet program such as Excel and use it to generate a graph To Save the Model Graph Click the L Save button A dialog box appears allowing you to select the file properties Type a name for your file Select the destination folder Click OK The plot is saved as a PNG formatted image To Print the Model Graph i Click the Print button situated under the title A dialog box appears allowing you to select the printer to use Select the printer to use and set other print properties if need be Click OK The report is printed To Export the Model Graph to Microsoft Excel Click the El Export to Excel button situated under the title An Excel sheet opens containing the model graph you are currently viewing along with its data A B C D E F G H J K Performance percentage 10 0 30 0 30 1 00 Performance 11 0 35 0 35 1 00 12 0 40 0 40 1 00 13 0 45 0 45 1 00 14 0 50 0 50 1 00 15 0 55 0 55 1 00 16 0 60 0 60 1 00 17 0 65 0 65 1 00 18 0 70 0 70 1 00 19 0 75 0 75 1 00 20 0 80 0 80 1 00 21 0 85 0 85 1 00 22 0 90 0 90 1 00 23 0 95 0 95 1 00 percentage ame andom Wizard alidation BF a WAH KeReportd Sheet fa CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved _ Infinitelnsight Modeler Segmentation Clustering l Note compatible with Excel 200
239. lities you want to add for example the two three or four best l Note When the SQL mode is activated the notion of nearest cluster does not exist If a case belongs to a cluster probability is set to 1 If a case does not belong to a cluster probability is set to O Outputs by Cluster Ranks Distance to Clusters This option allows you to add to the output file the distance of each observation from the clusters The distances are generated in the columns named kc_dist_cluster_ lt TargetVariable gt _ lt Clusterld gt For example if the target variable is Age the distance from cluster 1 will appear in the column kc_dist_cluster_Age_1 V To Add the Distances from All Clusters Check the All option VI To Select Distances from Specific Clusters 1 Check the ndividual option 2 Click the gt gt button to display the cluster selection table 3 Check the clusters for which you want to add the distance l Note When the SQL mode is activated the notion of nearest cluster does not exist If a case belongs to a cluster distance is set to O If a case does not belong to a cluster distance is set to 1 Probability for Clusters This option allows you to add to the output file the probability of each observation to belong to the various clusters The probabilities are generated in the columns kc_proba_cluster_ lt TargetVariable gt _ lt Clusterld gt For exmaple if the target variable is Age the probability that the observation
240. llection and extraction processes and tools and not SAP Infinitelnsight features On the other hand in order for your data to be usable by SAP Infinitelnsight the following five conditions must be met You must have a sufficiently large volume of data to be able to build a valid model that is in order for the model to be both relevant and robust An analytical model that is generated from a data set of 50 lines may have low generalization capacity and contain low informative value We can advise you on the issues of data volume Your data set must contain a target variable that will allow expression of your business issue within SAP Infinitelnsight The target variable must be known for each observation of the training data set To express this another way no target variable values may be missing over the range of the entire training data set The data source format must be supported by SAP Infinitelnsight Your data must be presented in the form of a single table of data except in instances where you are using the Infinitelnsight Explorer Event Logging or Infinitelnsight Explorer Sequence Coding features CUSTOMER SAP Infinitelnsight 7 0 15 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight 4 Essential Concepts This section introduces the essential concepts relating to use of SAP Infinitelnsight All concepts are introduced and appear in boldface in the sect
241. lusters or only the closest VI To Add All the Clusters Check the All option vV To Add Only the Closest Clusters 1 Check the Top option 2 Inthe text field enter the number of clusters you want to add for example the two three or four closest Top Ranking Centroids Names This option allows you to add to the output file the names of the clusters whose centroids are the closest to the current observation The closest cluster is the one the observation belongs to its name is displayed in the column kc_name_ lt Target variable gt The next closest cluster is displayed in the column kc_name_ lt Target Variable gt _2 and so on until the furthest cluster You can choose to add all the clusters or only the closest vV To Add All the Clusters Check the All option vV To Add Only the Closest Clusters 1 Check the Top option 2 Inthe text field enter the number of clusters you want to add for example the two three or four closest l Note The name of a cluster is by default its number you can modify this in the column User Name of the panel Clusters Profiles accessible through the main menu Top Ranking Distances This option allows you to add to the output file the distances of each observation from the clusters centroids The distance from the closest centroid is displayed int the column kc_best_dist_ lt TargetVariable gt the distance from the second closest centroid is displayed in the column kc_best_dist_ lt TargetVaria
242. mainly on the data set size and target distribution In versions prior to 6 1 0 Infinitelnsight automatically excluded variables with low prediction confidence since version 6 1 0 this behavior has been disabled by default If you do not enable this feature no variable will be excluded based on the value of its prediction confidence vV To Automatically Exclude Variables with Low Prediction Confidence Check the option Exclusion of Low Prediction Confidence Variables EEEN ture setectin Risk Mode Gain Chart wn eee eee en eee ia i m K ome ns Keep the First 1024 CUSTOMER SAP Infinitelnsight 7 0 100 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Defining the Number of Correlations to Display The section Correlations Settings allows you to set the parameters for the Correlation debriefing panel That is to select how many correlations should be displayed in that panel To say that variables are correlated implies that they each contribute some of the same information with respect to the target variable A correlation contains two variables and a correlation rate When you modify the number of correlations to display the engine excluded the ones with the lowest correlation rate thus keeping only the more significant ones Correlations Settings Higher than es 0 50 Keep all Correlations Keep the First 1 024 CUSTOME
243. model 264 CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Copy the Variables This option allows you to add to the output file one or more variables from the data set vV To Add All the Variables Check the All option vV To Select only Specific Variables Check the ndividual option Click the gt gt button to display the variable selection table In the Available list select the variables you want to add use the Ctrl key to select more than one variable Click the gt button to add the selected variables to the Selected list A WN User Defined Constant Outputs This option allows you to add to the output file constants such as the apply date the data set name or any other information useful for using the output file A user defined constant is made of the following information Parameter Description Value Warnings Visibility indicates if the constant will appear in the checked the constant appears in the output output or not unchecked the constant does not appear in the output Name the name of the user defined constant 1 The name cannot be the same as the name of an existing variable of the reference data set 2 Ifthe name is the same as an already existing user defined constant the new constant will replace the previous one Storage the constant type number string integer number date string
244. more information see the section Band Count for Continuous Variables In the section Grouped Cross Statistics with the Target s if the option Optimal Grouping is enabled the number of displayed categories is lower than that defined in the user structure by the parameter band count if no user structure has been defined For more information about using this parameter refer to the section Optimal Grouping for all Variables CUSTOMER SAP Infinitelnsight 7 0 137 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification the Model Performance in which you will find the model performance indicators the variables contributions and the score detailed statistics the Control for Deviations which allows you to check the deviations for each variable and each variable category between the validation and test data sets the Expert Debriefing in which you will find more specialized performance indicators as well as the variables encoding the excluded variables during model generation and the reason for exclusion and so on Variables Exclusion Cause Statistical Reports include the section Variables Exclusion Causes For regression and classification models this sections presents the reason why a variable got excluded from the model if any variable was It is divided in two parts Overall Exclusions showing the variables excluded from the whole model KXEN Inf
245. moved from this list and added to the list Variable from Loaded Model vV To View a Variable Structure Defined in the Loaded Model f the variable has not been added yet to the list of variables located on the lower part of the panel 1 Inthe list Variable from Loaded Model select the variable for which you want to see the structure defined in the model 2 Click the View button the variable structure opens in a new window PX 2r census 3390 Target from Loaded Model dass Variables from Training Census01 csv workdass Variables from Loaded Model K2R_Census_331_1 workdass Add __ X workclass marit Group Structure Category Edition ee I Federal gov X Federal gov JD KxMissing KxOther KxMissing KxOther Local gov Self emp not inc Local gov Self emp not inc d Private State gov Private V Enable the target based optimal grouping performed by K2C ce fthe variable has already been added to the list of variables located on the lower part of the panel Double click the variable for which you want to see the structure defined in the model CUSTOMER SAP Infinitelnsight 7 0 80 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification VI To Create or Modify a Variable Structure Double click the Structure icon ar corresponding to the variable for which you want to edit or create
246. mple for a target variable named class and a number of quantiles equal to 10 the generated column will be named quantile_rr_class_10 1 Check the option Predicted Value Quantiles 2 Inthe field Number of Quantiles enter the number of quantiles you want to create Contributions This option allows you to add the variables contributions for the current variable to the output file You can add the contributions of all variables or select only the contributions of specific variables It appears in the output file as contrib_ lt variable gt _rr_ lt target variable gt For example if marital status is an explanatory variable for the target variable class the column contrib_marital status_rr_class will be generated in the output file V To Add All Variables Contributions Check the All option vV To Add Specific Variable Contributions 1 Check the Individual option 2 Click the gt gt button to display the variable selection table 3 Inthe Available list select the variables you want to add use the Ctrl key to select more than one variable 4 Click the gt button to add the selected variables to the Selected list CUSTOMER SAP Infinitelnsight 7 0 172 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Types of Results Available The application of a model to a data set allows you to obtain four types of results which are described in the following table
247. n profession category and workclass socio professional category variables they must contain values During execution of the simulation SAP Infinitelnsight will automatically assign values to certain variables when values are missing but essential to proper completion of the simulation Once the simulation is complete you will obtain the following results The predicted value score The probability that this observation belongs to the target category of the target variable vV To Simulate a Model 1 Onthe screen Using the Model click the option Simulation The screen Simulating the Model appears 4 SAP InfiniteInsight VX y 2 class_Census01 File Help Simulating the Model Explanatory Variables Sort by Contribution of class kal Names Values Reset f Run marital status capital gain occupation education num age capital loss Variable hours per week Min education Max relationship SeX workclass native country a e aa Ga Ga E E Ea Ea a Ea LLL face E E E Results CUSTOMER SAP Infinitelnsight 7 0 177 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 2 On the left side of the screen Explanatory variables select a variable such as marital status Its values appears in the section Modifying values on the right side of the screen a SAP InfiniteInsight VX y 2 class_Census01 Fil
248. n Classification 182 On the bar Percentage of Information Retained move the cursor to change the amount of information to keep the number of variables selected changes accordingly The further this cursor is moved to the left the more variables are excluded The variables excluded are selected automatically as a function of their significance with respect to the model For instance the figure below shows that to retain only two variables out of the original fourteen you should keep 43 07 of the information contributed by the model Fa Smart Variables Selection ee a Percentage of Information Retained 43 32 Remaining Variables 2 Skipped Variables 12 Remark 0 variable s automatically exclud OK Cancel l Note Certain variables in the training data set may contribute no information such as constant value variables These can therefore be automatically excluded from the model during the training phase The number of variables excluded is displayed as a Remark In the figure above this number is equal to Oe CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 5 Click OK The window Smart Variables Selection closes and the panel Selecting Contributory Variables is updated with the selected variables allowing you to view the kept variables and the excluded ones In our example SAP Infinitelnsight automatically
249. n Confidence KR 0 996 Nb Variables Kept 714 44 Cancel 4 Previous a If the performance of the model meets your requirements go to Step 3 Analyzing and Understanding the Model Generated see page 113 b Otherwise go to the procedure To Generate a New Model CUSTOMER SAP Infinitelnsight 7 0 112 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 2 Youcanalso check other indicators provided in addition to KI and KR during the model generation For example you could view the total elapsed time required to generate the model and information on the standard error rate vV To Generate a New Model You have two options On the screen Training the Model you can Either click the Previous button to return to the modeling parameters defined initially Then you can modify the parameters one by one Or click the Cancel button to return to the main screen of Infinitelnsight modeling assistant Then you must redefine all the modeling parameters 5 2 3 Step 3 Analyzing and Understanding the Model Generated The suite of plotting tools within SAP Infinitelnsight allows you to analyze and understand the model generated The performance of the model with respect to a hypothetical perfect model and a random type of model The contribution of each of the explanatory variables with respect to the target variable The significance of the various categori
250. ndition 1 Click the button Add a Condition The window Define a Condition opens i Define a Condition on e 2 Choose a variable in the first list 3 Choose an operator in the second list 4 Indicate a value in the third list Fora variable with number storage type a value For a variable with string storage choose a variable in the list If the list is empty click the button to extract the variable categories CUSTOMER SAP Infinitelnsight 7 0 214 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 5 Click OK l Note You can edit a condition by double clicking it vV To Add a Logical Conjunction Click the button Add Logic And or the button Add Logic Or l Note You can change a conjunction by double clicking it vV To Change the Order You can change the order of the nodes to accelerate the filtering process by setting the conditions with the highest probability to be false at the top of the list 1 Select the node you want to move up or down 2 Usethebuttons and to change its position in the filter vV To Delete a Node 1 Select the node you want to delete 2 Click the button Remove Selected Node vV To Display the Filtered Data Set You can visualize the data set that you will obtain after the application of the filter Click the button View Data A pop up window opens CUSTOMER SAP Infinitelnsight 7 0 215
251. nfinitelnsight feature with which you will work and help you at all stages of the modeling process vV To Start Infinitelnsight modeling assistant 1 Select Start gt Programs gt SAP Business Intelligence gt SAP SAP Infinitelnsight gt Infinitelnsight modeling assistant Infinitelnsight modeling assistant screen appears fe SAP InfiniteInsight Vx y z File Help Modeler Welcome to Socia SAP Infinitelnsight Recommendation Toolkit Explorer Create or Edit Explorer Objects Create a Data Manipulation Load an Existing Data Manipulation Ferform an Event Log Aggregation Perform a Sequence Analysis Perform a Text Analysis 2 Click the feature you want to use Editing the Options vV To Edit the Options of Infinitelnsight modeling assistant In the File menu click Preferences The window Edit Options appears CUSTOMER SAP Infinitelnsight 7 0 62 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification The following options can be modified Category Options General Country Language Message Level Log Maximum Size Message Level for Strange Values Display the Parameter Tree Number of Store in the History Always Exit without Prompt Include Test in Default Cutting Strategy Stores Default Store for Apply in Data Set Default Store for Apply out Data Set Default Store to Save Models
252. ng the Model click Clusters Summary The panel Clusters Summary appears 4 SAP InfiniteInsight VX Yy Z cass_Census01 File Help alli Clusters Summary a T Asa a A View Type Bar Cluster Names Copy Print Save E portto Pin View Orientation Excel Boe Frequency Data Set Estimation a7 5 hours perweek S703 4 5 FR 7 Rec i0 i11 CUSTOMER SAP Infinitelnsight 7 0 247 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 2 Use the options to define the variables you want to display in the bubble chart The table below lists the available options The option allows you to Note that select the variable to be used in the X Axis only continuous and nominal numerical variables can be used select the variable to be used in the Y Axis only continuous and nominal numerical variables can be used select the variable to be used for the bubble only the variable Frequency and the size target variable can be used display cluster names cluster names can be customized in Cluster Profiles Understanding the Bubble Charts The bubble charts allow you to display the clusters representing the relationship between three variables Thus a bubble chart can provide 3 pieces of information on each cluster In addition the bubble charts provide a graphical representation of the segmentation enabling you to easily visualize the clust
253. nging the name of your style sheet The previous one is not deleted CUSTOMER SAP Infinitelnsight 7 0 64 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification VI To Delete a Style Sheet 1 Select one of the displayed style sheets 2 Clickthe button Remove Note that the style sheet is not only deleted from the list but also from the data source V To Edit the General Settings Settings Options Note Reports Background Color choose a color Only the PDF and HTML formats can display a make transparent background color Edit Configuration font size Check the option Dynamically render option font style changes or click Apply when editing the settings font color so that you can visualize the result text background color table configuration The selected settings will be applied to both the wizard and the generated reports vl To Edit the Charts Settings Settings Options Note Chart Colors modify the charts colors Default Chart Bars Orientation horizontal It is possible to set another default vertical orientation for specific report items vV To Edit Report Items 1 Setthe properties of your choice 2 Click Save to validate A window opens indicating that your style sheet has been successfully saved 3 Click OK Properties Functions Displayed as name of the label View Type choose between Tabular HTML and Gra
254. ngs Whole Population Population 48842 Positive Target B3 23 marital status marital status marital status marital status TNevwermansd Marnied AF spouse M Divorced Manned sp Seperated Widowed Population 16117 Population 22416 Population T251 Population 2048 Positive Target 4 55 Positive Target BEE c Positive Target 10 04 Positive Target 7 45 beei Profit Curve Detected h Estimation Validation Population Count 12461 48842 Positive Target Count 2973 11687 Positive Target Ratio 23 86 23 93 Negative Target Count 9468 37155 Negative Target Ratio 76 14 T607 Variance 0 02 Weighted Population 124561 0 CUSTOMER SAP Infinitelnsight 7 0 148 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Tne Decision Iree Each node in the tree displays the name of the expanded variable for example Marital status the categories on which the node population has been filtered for example Married AF spouse Never married the Population of the node the ratio of Positive Target for nominal targets or the Target Mean for continuous targets Example for Example for a nominal target a continuous target marital status workclass Married 4F spouse M Federal gov Local g Population 22416 Population 12106 Positive Target 44 Target Mean 43 172 When you go over a node several options are offered ro Selec
255. nk asks you to build a segmentation model of the customers of this product Using Infinitelnsight Modeler Segmentation Clustering you can rapidly develop a descriptive model with the least possible cost This model shows the characteristic profiles of the customers interested in your new product and thus responds to your business issue and fulfills your objectives 6 1 1 Presentation This scenario develops logically from scenario 1 In scenario 1 using the Infinitelnsight Modeler Regression Classification formerly known as K2R feature you managed to accomplish all the objectives of your first marketing campaign meeting the deadlines and within the budget you were allowed In order to customize the marketing messages from the bank and improve communication with the various customers and prospects for this new product the senior management of the bank now asks you to build a segmentation model of the customers of this product Using the Infinitelnsight Modeler Segmentation Clustering formerly known as K2S feature you can rapidly develop a descriptive model with the least possible cost This model shows the characteristic profiles of the customers interested in your new product and thus responds to your business issue and fulfills your objectives CUSTOMER SAP Infinitelnsight 7 0 197 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 6 1 2 Your Objectiv
256. ntributions This option allows you to add the variables contributions for the current variable to the output file You can add the contributions of all variables or select only the contributions of specific variables It appears in the output file as contrib_ lt variable gt _rr_ lt target variable gt For example if marital status is an explanatory variable for the target variable class the column contrib_marital status_rr_class will be generated in the output file V To Add All Variables Contributions Check the All option V To Add Specific Variable Contributions 1 Check the Individual option 2 Click the gt gt button to display the variable selection table 3 Inthe Available list select the variables you want to add use the Ctrl key to select more than one variable 4 Click the gt button to add the selected variables to the Selected list Nominal larget Outputs by Rank Scores This option allows you to generate in the output file the best score s for each observation For each line in the application data set SAP Infinitelnsight compares the scores obtained by the current observation for each category of the target variable and displays the best score in the column best_rr_ lt Target Variable gt _1 then if several scores have been requested the second best score is displayed in the column best_rr_ lt Target Variable gt _2 the third best in the column best_rr_ lt Target Variable gt _3 and so on When using this
257. nu allows the selection and graphing of any of the variables in the model The tool bar located under the title allows the user to copy the coordinates to the clipboard print the plot or save it in PNG format The values are normalized and their sum always equals to O0 Depending on the chosen profit strategy or on the continuous target variables value type you can obtain all positive importances or negative and positive importances CUSTOMER SAP Infinitelnsight 7 0 133 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Axes The X axis shows the influence of the variable categories on the target The significance of the different numbers on the X axis are detailed in the following table Number on the X axis Indicates that the category has positive number a positive influence on the target O no influence on the target the behavior is the same as the average behavior of the whole population negative number a negative influence on the target The Y axis displays the variable categories Categories sharing the same effect on the target variable are grouped They appear as follow Category_a Category_b Category_c Categories not containing sufficient numbers to provide robust information are grouped inthe KxOther category When a variable is associated with too many missing values the missing values are grouped in the KxMissing category Both categories are crea
258. ny All rights reserved Essential Concepts synonyms Depending upon your profile and your area of expertise you may be more familiar with one of the following terms to refer to explanatory variables Causal variables Independent variables nput variables These terms are synonyms Example Your company is marketing two products A and B You have a database which contains references to 1 500 of your customers You know which product A or B each customer has purchased 10 000 prospects You want to know which product each prospect is likely to purchase The variables name age address and socio occupational class are your explanatory variables they allow you to generate a model capable of explaining and predicting the value of the target variable product purchased The following table represents your database Name Age _ Adress Socio Occupational Class Product Purchased Charles 34 New Orleans Manager Administrator Product A John 37 Washington Manager Administrator Product A Marlene 31 Boston Civil servant Product B Prospect 1 34 Oakland Manager Administrator Prospect 2 24 Washington Civil servant Prospect n 35 Sacramento Skilled tradesman CUSTOMER SAP Infinitelnsight 7 0 33 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Weight Variable Definition A weight variable allows one to assign a relative weight to each of the observations it describes
259. o apply segments SQL expressions are built to describe as much as possible the basic segments that is the ones you get when you do not ask for SQL The SQL can be used both to have a better definition understanding of the clusters and to deploy them on the full database or on new data which is not usually trivial with other techniques The best way to understand the difference between centroid based clusters and rule based clusters is to use graphs Diagram Explanation This diagram represents a set of observations from a data set To create clusters Infinitel nsight Modeler Segmentation Clustering engine uses the centroid approach Centroids are the results of a clustering algorithm meaning they are the barycenter of the points closest to them When applying Infinitelnsight Modeler Segmentation Clustering on this data set the observations are grouped depending on their distance with each centroid This graph represents the previous data set observations grouped into four clusters This diagram is known as a Voronoi diagram To create the SQL expressions that define the clusters Infinitel nsight Modeler Segmentation Clustering engine uses what is called Minimum Description Length MDL It means that after creating the initial clusters from the centroid approach then they are reshaped cut to fit into the smallest possible expression thus trying to find the best compromise between length of the expression and the lost of info
260. o be false at the top of the list 1 Select the node you want to move up or down 2 Usethebuttons and to change Its position in the filter vV To Delete a Node 1 Select the node you want to delete 2 Click the button Remove Selected Node vV To Display the Filtered Data Set You can visualize the data set that you will obtain after the application of the filter Click the button View Data A pop up window opens a InfiniteInsight Sample Data View Data Set First Row Index 1 Last Row index 100 fe Census01 csyv atic Statistics Graph i 90 age workclass fntwat education education marital st occupation relations 1 25 Sel emp n 176756 HS grad 9 Never marr Farming fis Qwn child 2 25 Private 269980 HS grad 9 Never marr Handlers c Not in far 3 25 Private 32275 Some colle 10 Married civ Exec mana Wife 4 25 200681 Some colle 10 Never marr Own child 5 25 Private 252752 HS grad 9 Never marr Other servi Unmarrie 6 25 Private 255004 10th 6 Never marr Crafi repair Not in fan T 25 Frivate 159732 50me colle 10 Never marr Adm clerical Not in fan 8 25 Private 193787 Some colle 10 Never marr Tech supp Qwn child 9 25 Private 371987 Bachelors 13 Never marr Exec mana Mot in fan 10 25 Private 344991 Some colle 10 Married civ Craft repair Husband 11 25 Private 86872 Bachelors 13 Married civ Exec mana Husband 12 25 Private
261. odel 4 SAP InfiniteInsight VX y 2 class_Census01 File Help 3 View Source Code awk a gt a Copy Print Save SAP Infinitelnsight SAP II 6 5 SP5 Temporary License File Evaluation 7 0 0 q13 Copyright 2014 SAP AG or an SAP affiliate company All rights reserved Model builtin 7 0 0 913 Model Name is class Census01 Model Version is 1 function fabs a CONVEMT 56 179 iff a gt 0 return a else return a I function doublecmpt a b CONVEMT 90 179 if a b return 0 else return 1 function doublesegcmp ix iXStart iEqualStart iMEnd iEqualEnd CONVEMT 56 179 if iM iMStart return 2 if X iXStart if iEqualStart return 0 else return 1 CUSTOMER SAP Infinitelnsight 7 0 187 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification List of Generated Codes The following table lists the available codes with their particularities Generated Code Comment C see C Code Generator documentation JAVA needs the Kx RT jar package to run PMML3 2 AWK CPP SAS SQLServer wraps variable names with SQLServerUDF HANA Infinitel nsight Scorer manages HANA column and row storage ORACLE OracleUDF SQLDB2 DB2UDF DB2V9 SQLTeradata Teradata databases TERAUDF MYSQL MYSQLUDF SybaselQ Sybasel QUDF SQLNetezza SQLVertica PostgreSQL Geenplum Hive HTML Javascrip
262. omatically sort count and total the data stored in one table or spreadsheet and create a second table displaying the summarized data Pivot tables are also useful for quickly creating cross tabs polynomial A polynomial may be of degree 1 2 3 or greater By defining the polynomial degree you are defining the degree of complexity of the model population A population is a list of entity identifiers A population may be defined as list of values This list can be extracted from a column table it is then said to be defined in extension or through a filtering expression from another population it is then said to be defined in intension prediction range The extreme values for prediction ranges are TargetMean sqrt TargetVariance TargetMean sqrt TargetVariance predictive model A model which allows predicting phenomena CUSTOMER SAP Infinitelnsight 7 0 285 2014SAP AG or an SAP affiliate company All rights reserved Glossary profit type A profit type allows calculation of the profit that may be realized using the model In general a benefit is associated with the positive or expected values of the target variable and a cost is associated with the negative or unexpected values Q quality indicator the predictive power KI The predictive power is the quality indicator of the models generated using SAP Infinitelnsight This indicator corresponds to the proportion of information contained in the target
263. on belongs The variable kc_TargetWMeanClusterld which indicates the proportion of observations belonging to the target category of the target variable that are contained in each cluster The variables corresponding to each cluster and an indication of the encoding disjunction of the cluster numbers The names of these variables correspond to cluster numbers prefixed by kc_cluster_ for example kc_cluster_1 for cluster 1 CUSTOMER SAP Infinitelnsight 7 0 270 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 7 Glossary analytical data management The analytical data management as defined by SAP Infinitelnsight is made of three elements the data manipulation functions offered by SAP Infinitelnsight such as filters join attributes new attributes see attribute on page 272 computation aggregates see page 2 6 performance indicators see page 284 definition the SAP Infinitelnsight analytical data set methodology the meta data management which allows storing sharing and easily re using the data descriptions analytical data set Tabular representation of data made of lines and columns Each line represents an observation Roles see page 287 can be assigned to columns such as Input skip target or weight analytical record An analytical record is a logical view of all attributes see attribute on page 2
264. on to display the curve chart The curve plot appears a SAP InfiniteInsight VX y 2 cass_Census01 File Help ali Category Significance A a Bbg w Data Sets View Type Bar Copy Print Save EE to Pin View H i Ki Sa es Orientation r Variables age ha Variable age Detected Profit ED gD aD gD gD gD e a gho ah am BS BS th fh ath gts a ath gh aah AAS AT aT aA AT AC AO AO gA OO CO A SU Categories E Fandom E Wizard Estimation F Validation 2 Click View Type and select the ally button to go back to the bar chart l Note You can combine the different types of plot For example you can display All Datasets in a curve chart or the Validation Data Set in a bar chart CUSTOMER SAP Infinitelnsight 7 0 242 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Understanding the Plots of Variables For this Scenario Select the variable marital status which is the explanatory variable that contributes the most to the target variable Class a SAP InfiniteInsight VX y 2 cass_Census01 File Help ally Category Significance A E 9 Ams A Data Sets View Type Bar Copy Print Save EXPOrttO pin View ae tet eee Orientation r Excel Variables marital status w Variable marital status Influence an Target 0 3 0 2 0 1 0 0 0 1 0 2 0 3 0 4 0 5 Married AF spouse Married civ spouse Divorced Maried
265. ons 1 501 to 3 000 The third file observations 3 001 to 5 000 A Caution The customized cutting strategy is risky in the instance of an initial data file in which the data have been sorted In this case the first lines will not be representative of the overall set of data contained in the first file To avoid this type of bias do not forget to mix up your data prior to analysis seven Automatic Cutting Strategies Background With the exception of the customized cutting strategy cutting strategies are automatic Automatic cutting strategies operate upon a single data file which constitutes your initial data set Automatic cutting strategies always cut the initial data set into the same proportions The following table details the proportions attributed to each data set depending on the presence of a test data set Automatic Cutting Strategies with Test Automatic Cutting Strategies without Test 3 5 of the data are used in the estimation sub set 3 4 of the data are used in the estimation sub set 1 5 of the data are used in the validation sub set 1 4 of the data are used in the validation sub set 1 5 of the data are used in the test sub set 4 reel 1 ESMER 2 Validation T5 25 CUSTOMER SAP Infinitelnsight 7 0 21 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Random The Random cutting strategy distributes the data of the initial data set in a random manner between the thr
266. ontinuous 0 0 a 4 education string nominal if 0 z 5 education n number ordinal 0 0 a 6 marital status string nominal aa T occupation string nominal 0 0 a Srelationship string nominal 0 0 a grace string nominal 0 0 aa 10 sex string nominal 0 0 a 11 capital gain number continuous 0 0 99999 a 12 capitalloss number continuous 0 0 ae 13 hours per w number continuous 0 0 a 14 native country string nominal 0 0 a 15 class number nominal 0 0 ae 16 Kxlndex integer continuous 14 0 Automaticall aa _ Add Filter in Data Set cs Om 6 Click Next Why Describe the Data Selected In order for Infinitelnsight features to interpret and analyze your data they must be described To put it another way the description file must specify the nature of each variable determining their Storage format number number character string string date and time datetime or date date Type continuous nominal ordinal or textual For more information on data description go to Types of Variables on page 26 and Storage Formats on page 29 CUSTOMER SAP Infinitelnsight 7 0 21l 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering How to Describe Selected Data To describe your data you can Either use an existing description file that is taken from your information system or a previously created description file from Infinitelnsight feature
267. ory variables they allow calculation of the value of the target variable in a given context They may also be used as weight variables For more information about the role of each SAP Infinitelnsight features see Operations see page 12 You can then generate models on page 35 capable of either explaining and predicting a phenomenon or describing a data set in both cases as a function of the previously defined target variable This phase is called the training phase Once the models have been generated you can view and interpret their relevance and robustness using Performance indicators on page 39 the predictive power which is the quality indicator and the prediction confidence which is the robustness indicator A variety of plots including the profit curve plot CUSTOMER SAP Infinitelnsight 7 0 17 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 2 Data Sources Supported In the standard version SAP Infinitelnsight supports the following data sources Flat files text files in which the data are separated by a delimiter such as commas in csv Comma Separated Value format file For instance the sample file CensusOl csv used for the Infinitelnsight Modeler Regression Classification and Infinitelnsight Modeler Segmentation Clustering application scenarios is a csv file _ODBC compatible data sources if your license allows it you can also use SAS files
268. oss of predictive power Where possible similar adjacent segments are gathered to reduce artifacts between the estimation and validation data sets v To Enable Infinitelnsight Modeler Data Encoding Optimal Grouping for All 1 2 88 Variables Right click the row corresponding to the variable to be edited Select Define Structure rf SAP InfiniteInsight Vx y z7 New Regression Classification Model File Help Data Description Main Edition Structures fe New Structure Edit User Band Count From Statistics From Model Y Remove Structure From Variable Description Desc_Census01 csv Index Name Storage Value Key Order Missing oo Storage 0 0 Value rious lO v Missing Values 0 0 a Translate categories 0 0 n 0 LI Define Structure From Statistics Encoding j For Selected Variables Save in Variable Fool Far aiher i Properties From Variable Fill Groups From Domains New Structure Set Group Remove Structure Edit User Band Count 14 native country string nomin 15 class number nomin Optimal Grouping 16 Kxlndex integer continuous 1 0 Add Filter in Data Set 44 Cancel CUSTOMER Group Description Structure Automaticall co Om SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 3 Select Optimal Grouping in such a way that the option is checked Filtering the
269. ots with the prefix c_ For example the encoded version of a continuous variable named age is noted c_age l Note In Infinitelnsight Modeler on the Data Description panel if you enable the Natural Encoding for a given variable its K2C encoded value c_variableName will not be generated Category Significance Definition The Significance of Categories plot illustrates the relative significance of the different categories of a given variable with respect to the target variable CUSTOMER SAP Infinitelnsight 7 0 128 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Displaying the Significance of Categories Plot vV To Display the Significance of Categories Plot 1 On the screen Using the Model click Category Significance The plot Category Significance appears a SAP InfiniteInsight VX y 2 class_Census01 File Help ali Category Significance A wg 0 DataSets ViewType Bar T T i add at oc Export to Copy Print Save Excel r Pin View Variables age vw Variable age Influence on Target 0 075 0 050 0 025 0 000 0 025 0 050 0 075 0 100 0 125 0 150 0 175 0 200 categories E Yalidation 2 Inthe Variables list located above the plot select the variable for which you want to display the categories If your data set contains date or datetime variables automatically generated variables can appear in the V
270. ound minus Its lower bound CUSTOMER SAP Infinitelnsight 7 0 49 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Density Good This curve displays the distribution of model scores for responders signals Performance 2 00 Density 46 29 25 aP pip vob a ab nm i aai al et ne q om pae 26 ono od ge i Beh fe DB score E Random Validation Density Bad This curve displays the distribution of model scores for non responders non signals Performance 2 25 2 00 1 75 1 50 1 25 1 00 Density 0 75 0 50 0 25 0 00 AD a a gt yah a D ook nO al ool El nee pA oE oom cat q0 pita ae a 22 s Pa at ah a ah ata ar oF oF oF OF ah gigi ge OT A score E Random Validation CUSTOMER SAP Infinitelnsight 7 0 50 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Density AIl This curve displays both the curves Density Good and Density Bad thus allowing the user to compare both distributions Performance 2 25 2 00 1 75 1 50 1 25 Density 1 00 0 75 0 00 NO yD oF 4D agai aa e nh a aa a aa pi y ne wal ol ool ye pl on oo eh oh gh gg ge gg gS ye alia OF A Score BRandom ValidationDensity Bad ValidationDensity Good 4 10 4 Risk Curves Good Bad Odds The X axis represents the risk score and the Y axis represents the odds ratio
271. phase of your marketing campaign CUSTOMER SAP Infinitelnsight 7 0 248 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Displaying Cluster Plots vV To Display Cluster Plots 1 Onthe screen Using the Model click Clusters Summary The panel Clusters Summary appears 2 Click the button A View Type and select Bar Chart A SAP InfiniteInsight VX y 2 cass_Census01 File Help aly Clusters Summary wa 9 ia Asa i View Type Bar Cluster Names Copy Print Save ExXportto Pin View z p Orientation Excel w Flot Relative Target Means Descending Sort Relative Target Means Data Set Estimation Clusters E Relative Target Means 3 Inthe Plot list select the type of plot that you want to display l Note Select the option Descending sort to sort the plot bars in descending order For instance on the plot Relative Target Means the descending sort allows quick examination of the most interesting clusters that is those which differ most from the mean behavior of the data set taken as a whole CUSTOMER SAP Infinitelnsight 7 0 249 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Understanding Cluster Plots The Plot Target Means The Target Means plot presents the proportion of observations belonging to the target category of the target variable present
272. phical The last one is only available if the report item can be displayed as a graph Chart Type select one of the proposed chart types Note that this option is only available for report items of the view type Graphical Switch Bar Orientation this option allows having another bar orientation as the default one for a specific report item Sort by Sort Order you can select a column to sort by and choose between an ascending or a descending order Visibility you can hide columns of a report item or even menu items Note that at least one column of a report item must remain visible vV To Apply the New Style Sheet to the Generated Reports 4 Inthe panel Report select the new style sheet 5 ClickOK A window opens indicating that you have to restart the modeling assistant to take the edited options into account 6 ClickOK When training a model all the generated reports the learn excel statistical reports are now customized CUSTOMER SAP Infinitelnsight 7 0 65 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Defining a Metadata Repository The metadata repository allows you to specify the location where the metadata should be stored vV To Define a Metadata Repository 1 Choose between storing the metadata in the same place as the data or in a single place by checking the option of your choice 2 Inthe list Data Type select the type of data you want to access
273. ponds to the reaction of prospects contacted during the test You assigned The value 1 for those prospects who responded positively to your invitation The value O for those prospects who responded negatively to your invitation VI To Select a Data Source 1 On the screen Select a Data Source select the data source format to be used Text files Data Base fe SAP InfiniteInsight Vx y z File Help Select a Data Source a Use a File or a Database Table Use Explorer Data Type Text Files Folder U U Samples Census Metadata are stored in the same place as data source 44 Cancel EJ You must select a data set 41 Previous I Next CUSTOMER SAP Infinitelnsight 7 0 208 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 2 Click the Browse button The following selection dialog box opens Select Source Folder for Data oF fi gd BG C Usersinatacha yam Documents Ay A HE Census H JapaneseData H El KelData H E KSN E E KTS l Ci i i i i g JA KA E PA Tl Ti i Tl i ld Samples ext Files dat data csv tt Use r Passwo rd oK Cancel 3 Double click the Samples folder then the Census folder l Note Depending on your environment the Samples folder may or may not appear directly at the root of the list of folders If you selected the default settings
274. presents the ratio between a model and the random model that is the performance of a model compared to a model that would only allow to select observations at random from your database You can thus visualize how much better your model is compared with the random model CUSTOMER SAP Infinitelnsight 7 0 46 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Standardized Profit Standardized profit allows examination of the contribution of the model generated by SAP Infinitelnsight features relative to a model of random type that is in comparison with a model that would only allow selecting observations at random from your database This profit is used for the plots of variable details which present the significance of each of the categories of a given variable with respect to the target variable Customized Profit Customized profit allows you to define your own profit values that is to associate both a cost and a benefit to each value of the target variable For instance you can define the cost of sending out a mailing as a negative value for example 5 the benefit brought in by the response to that mailing as a positive value for example 20 4 10 Advanced Model Curves In addition to the profit curves detailed in the previous section a series of advanced model curves are available in SAP Infinitelnsight 4 10 1 ROC The ROC Receiver Operating Characteristic graph is derived f
275. prospects to contact on a basis of real returns It is true that you probably have a relatively good understanding of which individuals stand a good chance of becoming your customers some day But optimizing your campaign means being able to identify those prospects that have every chance of becoming customers today as a result of the current marketing campaign Discover new niche prospects that all your knowledge of the market had not previously allowed you to identity Select a predefined number of individuals Imagine that one of the constraints of your campaign consisted of contacting exactly 5 000 prospects Your intuition may help you to select 2 400 of these But how are you going to identify the remaining 2 600 prospects to be contacted A purely random selection thus completely non optimized might be your only solution CUSTOMER SAP Infinitelnsight 7 0 58 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Classical Statistical Method You may decide to use a classical statistical method to better manage the effectiveness of your campaign and thus your budget On the basis of the information that you have a Data Mining expert could create predictive models In other words you could ask a statistical expert to create a mathematical model that would allow you to predict the probability of a given individual to respond to your marketing campaign as a function of
276. r by the title of the degree earned Number of years of study represented by a numerical value Marital status Job classification Position in family Ethnicity Gender Annual capital gains Annual capital losses Country of origin Variable indicating whether or not the Salary of the individual is greater or less than 50 000 Example of Values Any numerical value greater than 17 Private Self employed not inc Any numerical value such as 0 2341 or 205019 lith Bachelors A numerical value between 1 and 16 Divorced Never married Sales Handlers cleaners Husband Wife White Black Male Female Any numerical value Any numerical value United States France 1 if the individual has a salary of greater than 50 000 0 if the individual has a salary of less than 50 000 In order to avoid complicating the Infinitelnsight Modeler Regression Classification and Infinitelnsight Modeler Segmentation Clustering application scenarios the variable fn wgt is used as aregular explanatory variable in these scenarios and not as a weight variable CUSTOMER 61 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification SAP Infinitelnsight 7 0 5 1 8 Infinitelnsight modeling assistant To accomplish the scenario you will use the Java based graphical interface of SAP Infinitelnsight This interface allows you to select the SAP I
277. r field select the folder where the description file is located with the Browse button l Note The folder selected by default is the same as the one you selected on the screen Data to be Modeled In the Description field select the file containing the data set description with the Browse button A Caution When the space used for model training contains a physical variable named kxIndex it is not possible to use a description file without any key for the described space When the space used for model training does not contain a physical variable named kxIndex it is not possible to use a description file including a description about a KxIndex variable since it does not exist in Current space CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 5 Click OK The window Load a Description closes and the description is displayed on the screen Data Description 4 SAP InfiniteInsight Vx y Z New Regression Classification Model File Help Data Description A Open Description J Save in Variable Pool A i Analyze ke Save Description Remove from Variable Pool View Data Properties Description Wi VIEW Description Desc_Census01 csv Index Name Storage Value Key Order Missing Group Description Structure lage number continuous 0 a 2 workclass string nominal 0 0 l i 3 fniwat number c
278. r greater than 0 98 is very robust It has a high capacity for generalization Less than 0 95 must be considered with caution Applying it to a new data set will incur the risk of generating unreliable results Improving the Prediction Confidence of a Model To improve the prediction confidence of a model additional observation rows may be added to the training data set Predictive Power Prediction Confidence and Model Graphs On the model graph plot Of the estimation data set default plot the predictive power corresponds to the area found between the curve of the model generated and that of the random model divided by the area found between the curve of the perfect model and that of the random model As the curve of the generated model approaches the curve of the perfect model the value of the predictive power approaches 1 Of the estimation validation and test data sets select the corresponding option from the list Data set located below the plot the prediction confidence corresponds to one minus the area found between the curve of the estimation data set and that of the validation data set divided by the area found between the curve of the perfect model and that of the random model CUSTOMER SAP Infinitelnsight 7 0 AO 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Advanced Users Predictive Power for Continuous Targets 1 Working with the Validation data set use a uniform
279. r more Target Variables on page 31 possibly a Weight Variable on page 34 the Explanatory Variables on page 32 CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Target Variables For this Scenario Select the variable Class as your target variable that is the variable that indicates the probability of an individual responding in a positive or negative manner to your campaign To Select Targets Variables 1 Onthe screen Selecting Variables in the section Explanatory variables selected left hand side select the variables you want to use as Target Variables fe SAP InfiniteInsight Vx y z New Regression Classification Model File Help Selecting Variables Explanatory Variables Selected 14 Target Variables 1 workclass fniwgt education education num marital status occupation relationship og face SEX _ Alphabetic Sort capital gain Weight Variable 0 capitalloss hours per week ane Excluded Variables ff Aw H _ Alphabetic Sort M Alphabetic Sort l Note On the screen Selecting Variables variables are presented in the same order as that in which they appear in the table of data To sort them alphabetically select the option Alphabetic sort presented beneath each of the variables list 2 Click the button gt located on the left of the screen section Target Variables upper right hand side
280. ression Classification 3 Click the Browse button 195 A selection dialog box appears Select Source Folder for Data ay l g E H a i BECEARB BEBEasa HTA H Census J JapaneseData KAR KelData KSN KTC KTS Ti Us dd Samples Text Files dat data csv tct ha Use r Fasswo ra OK Cancel CUSTOMER 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification SAP Infinitelnsight 7 0 4 Select the folder that holds the model that you want to open The list of models contained in that folder appears providing the following information for each model Column Description Values Name Name under which the model has Character string been saved Class Class of the model that is the Kxen Classification Classification Regression with nominal type of the model target Kxen Regression Classification Regression with continuous target Kxen Segmentation Clustering with SQL Mode Kxen Clustering Clustering without SQL Mode Kxen TimeSeries Time Series Kxen AssociationRules Association Rules Kxen SimpleModel Classification Regression and Clustering multi target models any other model Version Number of the model version Integer starting at 1 when the model has been saved several times Date Date when the model has been Date and time in the format yyyy mm dd hh mm ss saved Commen Option
281. rkclass 0 004575992 0 004575992 0 002210456 0 002210456 0 0022160456 0 002210456 0 002216456 0 0045 5992 0 002216456 0 002210456 0 002216456 0 0045 5992 0 002216456 0 002210456 0 002216456 Depending upon the format of the results file generated use Microsoft Excel or another application to open the file The figure below presents the headings and columns of the results file obtained for this scenario 2 You can now analyze the results obtained and use these results of your analysis to make the right decisions Description of tne Results File Depending upon which options you selected the results file will contain some or all of the following information in the same order as seen below 176 The key variable defined during data description at the setting model parameters step Possibly the target variable given as known values if the latter appeared in the application data set as is the case in this scenario The predicted value score provided by the model for the target variable of each observation The name of this column corresponds to the name of the target variable prefixed by rr_ or in this case rr_Class The decision is based on the score For example its value can be of 1 if the observation is considered as interesting or Oif it is considered as uninteresting for the model The name of this column corresponds to the name of the target variable prefixed by decision_rr_ or in
282. rmation This graph represents the SQL expressions of the clusters in red compared with the centroids You can see on this graph that some observations that were in a cluster when using the centroid approach end up in another when using the SQL expressions some observations can not be described by the SQL expressions and are left outside the cluster They are called the unassigned observations some observations are described by two different SQL expressions thus appearing in two clusters This is called the overlap This graph presents the final result obtained with SQL expressions An observation cannot appear in two different clusters so when there is overlap between i clusters the observation concerned by the overlap is kept in the first cluster created The second cluster that was also containing the observation is redefined to exclude it In this schema the numbers correspond to the order of creation of the clusters You can see that the observations that were in two clusters are kept in only one The choice of the cluster in which the overlapping observations are kept depends on the order in which R the SQL rules are applied In this case the rule defining cluster 2 has been applied before the rules defining the clusters 1 and 3 CUSTOMER SAP Infinitelnsight 7 0 260 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Diagram Explanation
283. rom all the graphs and their attributes Infinitel nsight Social models only KxCommunities matches the nodes to their communities if the community detection was Infinitelnsight Social models enabled only A Caution When sharing or sending a model all these files must be joined to the model or the recipient will not be able to open the model CUSTOMER SAP Infinitelnsight 7 0 193 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Opening an Existing Model Once saved models may be opened and reused in SAP Infinitelnsight vV To Open a Model 1 Onthe main Infinitelnsight modeling assistant screen select Load a Model The screen Opening a Model appears fe AP InfiniteInsight Vx y z File Help Opening a Model Data Type Text Files ha Folder S5amplesiCensus Name Class Version Date Comment class Census01 Kxen Classification 1 2014 05 06 09 25 07 census class Census01 Kxen Classification 2 2014 05 06 11 21 30 census class Census01 Kxen Classification 3 2014 05 06 11 24 09 census Delete Selected 2 Inthe Data Type list select one of the following options Text files Database SAS files SAS Transport depending upon the format of the model that you want to open CUSTOMER SAP Infinitelnsight 7 0 194 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Reg
284. rom signal detection theory It portrays how well a model discriminates in terms of the tradeoff between sensitivity and specificity or in effect between correct and mistaken detection as the detection threshold is varied Performance Sensitivity O ob a4 a Og 45 h og ah og neg Ogi a 1 Specificity E Random E Wizard Validation ar 9759 y CUSTOMER SAP Infinitelnsight 7 0 47 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Sensitivity which appears on the Y axis is the proportion of CORRECTLY identified signals true positives found out of all true positives on the validation data set 1 Specificity which appears on the X axis is the proportion of INCORRECT assignments to the signal class false positives incurred out of all false positives on the validation dataset Specificity as opposed to 1 specificity is the proportion of CORRECT assignments to the class of NON SIGNALS true negatives 4 10 2 Lorenz Curves Lorenz Good Lorenz Good displays the cumulative proportion of missed signals false negatives accounted for by the records corresponding to the bottom x of model scores Performance o y 1 Sensitivity a ari w wiat age ted je ghe ca pigo ig ia ptp O79 4y percentage E Fandom E Wizard Validation The Y axis measures 1 sensitivity that is 1 the proportion of true positives which is equivalent to th
285. root of the mean of the quadratic errors Euclidian distance or root mean squared error RMSE Formula N TT T 2 Aha E y 1 V U fy i SSE N MSE ee W Wa L2 y4MSE Maximum Error LInf Definition maximum absolute difference between predicted and actual values upper bound Chebyshev distance Formula Lo maxu i Error Mean Definition mean of the difference between predictions and actual values Formula Mean Percent Error MPE l J Y MPE Y w 1 i l a Mean Absolute Percent Error MAPE y 3 F j l V CUSTOMER 44 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight 7 0 Essential Concepts Error Standard Deviation Definition dispersion of errors around the actual result Formula Classification Rate Definition ratio between the number of correctly classified records and the total number of records Formula GatHBbp t CR t E G E pea HL Papi Determination Coefficient R2 Definition ratio between the variability sum of squares of the prediction and the variability sum of squares of the data Formula N SSR Yw O 0 2 wO Z i N SST Nw 67 _ y p a i gt SSR SST CUSTOMER 45 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight 7 0 Essential Concepts 4 9 Profit Type 4 9 1Definition A profit type allo
286. rresponding equation is 1l AUC t dil 8 ly dy 00 0 CUSTOMER SAP Infinitelnsight 7 0 42 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts So we have AUC ama 1 1 27 x AUC mam fir ty 0 do One of the interest of the measure of AUC is its independence from the target distribution let us imagine that we build another data set where we duplicate each good example twice the AUC of the model will be the same A Caution Area Under the Roc Curve AUC has very nice properties to evaluate a binary classification system It is widely used now by statisticians even if it is not easy to picture for non statisticians 4 8 3 Error Indicators First some basic notations Target response value Vi Predictor predictor response value f a Residual 7 A i Error 4 1 y Ely Weight of the tested observation W W X w Total weight of the population i Af y a Sow Vi Target average W aay z 1S J DW Predictor average W a Mean Absolute Error L1 Definition mean of the absolute values of the differences between predictions and actual results City block distance or Manhattan distance Formula 1 Nv L X wu i CUSTOMER SAP Infinitelnsight 7 0 43 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Mean Square Error L2 Definition square
287. rt gt Programs gt SAP Business Intelligence gt SAP SAP Infinitelnsight gt Infinitelnsight modeling assistant Infinitelnsight modeling assistant screen appears fe SAP InfiniteInsight Vx y z File Help Modeler Welcome to Socia SAP Infinitelnsight Recommendation Toolkit Explorer Create or Edit Explorer Objects Create a Data Manipulation Load an Existing Data Manipulation Ferform an Event Log Aggregation Perform a Sequence Analysis Perform a Text Analysis 2 Click the feature you want to use Editing the Options vV To Edit the Options of Infinitelnsight modeling assistant In the File menu click Preferences The window Edit Options appears CUSTOMER SAP Infinitelnsight 7 0 203 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering The following options can be modified Category Options General Country Language Message Level Log Maximum Size Message Level for Strange Values Display the Parameter Tree Number of Store in the History Always Exit without Prompt Include Test in Default Cutting Strategy Stores Default Store for Apply in Data Set Default Store for Apply out Data Set Default Store to Save Models Metadata Repository Enable Single Metadata Repository Edit Variable Pool Content Graphic Profit Curve Points Bar Count Displayed No Infinitelnsight LoOK
288. ry Edition 10th 1st 4th di fAssoc acdmAssoc voc i Assoc acdm i ASSOC VOC di Bachelors i Bachelors di Masters i Masters Ji iDoctorate Prot school Add New Group Add Category i Doctorate Protschool New Category BW m m gt Advanced OK Cancel CUSTOMER SAP Infinitelnsight 7 0 84 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 2 Click the button Add New Group A group containing the selected categories is created in the list Group Structure education wt Structure Group Structure Category Edition J a ee SS oar a oS ed n I h Assoc acdm he ASSOCNOC J Bachelors i Bachelors J Masters i Masters di Doctorate Prof school i Doctorate i Prof school di H5 grad HS grad di isome college Some college Add New Group Add Category ist 4th Preschool i Preschool Add Missing _ Alphabetic Sort To Include Missing Values in a Group 1 Inthe list Group Structure select the group in which you want to add the missing values 2 Click the button Add Missing located under the list Category Edition The KxMissing category which represents the missing values is added to the selected group and the button Add Missing is deactivated As any category the KxMissing category can only belong to one group ata time g education wt Struct
289. s Or create a description file using the Analyze option from Infinitelnsight modeling assistant In this case you must validate the description file obtained You can save this file for later use A Caution The description file obtained using the Analyze option results from the analysis of the first 50 lines of the initial data file In order to avoid all bias we encourage you to randomly sort your data set outside SAP Infinitelnsight before performing this analysis Viewing the Data To help you validate the description when using the Analyze option you can display the first hundred lines of your data set Vl To View the Data 1 Click the button View Data A new window opens displaying the data set top lines a InfiniteInsight Sample Data View Data Set Cens usO1 csv First Row Index 1 Last Row index 100 E co Statistics Graph age workclass fntwat education education marital st occupation relations L 39 State gov T516 Bachelors 13 Never marr Adm clerical Not in far 2 50 Self emp n 83311 Bachelors 13 Married civ Exec mana Husband 3 36 Private 215646 HS grad 9 Divorced Handlers c Not in fan 4 3 Private 234721 11th T Married civ Handlers c Husband 5 26 Private 336409 Bachelors 13 Married civ Prof specia Wife 6 37 Private 284582 Masters 14 Married civ Exec mana Wife T 49 Private 160187 9th 5 Married sp Other servi Not in fan 8 Self emp n
290. s Each step removes Variables a Each step keeps 95 0 of information Search process stops with a drop of 5 0 of Predictive Power and the Prediction Confidence co oc CUSTOMER SAP Infinitelnsight 7 0 104 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification setting the Variables Auto sSelection The section Auto selection allows you to automatically reduce the number of variables in the model in relation to quality criteria This selection is done by successive iterations There are two selection modes one based on the number of variables to keep and the other on the amount of information that should be kept In this instance the information is the sum of the variables contributions vV To Use the Auto selection Check the box Enable Auto Selection The corresponding options are activated W Enable Auto selection Select the best model keeping between 1 and all variables By default the parameters are set to Se ectthe best model keeping between1 and all variables Any parameter that can be changed is marked as a hyperlink blue underlined Choosing the selection Mode VI To Select the Selection Mode 1 Click the link indicating the type of model to keep For example the best model in the sentence Se ectthe best model keeping between and all variables A drop down menu is displayed offering the following options the best model the last model 2
291. s Wilete 21 BG 7 av 12 Geena ume tee Peretti tree Meare et earn ores A A ee eee ne eee eee ee eee vodell GaBe ee nen eee E ane ane ate ee me ee eee ee eee ee ee eee eee ee Category SIGMMMC ANCES soeeccaacees cree cnetcedovergaesseessaqcuiacsetncecetuessacahensoapdedeenatoedeeneuioueeseesheei AEI E REA EAEri eii Clusters GUMMA Y ests etesccnacs chteiatnedeceedesnadenstesacsndectataceecmoseenesonchoneseacnetdstmsaededeadeensdeednedcensdnceseeatnedechodesaeiancienacess RMIT ON cas este a a ated ee ae E ete eased oceast E SATIS Call Repons esteen es ara A EEEE EERE E A EEEE EREE A suite of plotting tools allows you to analyze and understand the model generated The performance of the model with respect to a hypothetical perfect model and a random type of model The characteristics of each of the clusters The significance of the various categories of each variable of a cluster with respect to the target variable cross Statistics User Menu Once the model has been generated Click Next The screen Using the Model appears 4 SAP InfiniteInsight VX y 2 cass_Census01 File Help Using the Model Display ah Display Run Model Overview Model Graphs Save Export Category Significance Clusters Summary Cluster Profiles Statistical Reports Preis The screen Using the Model presents the various options for using the model that allow you to Display the information relating to the model generated Display section re
292. s 00 1Zhours per w number continuous 0 14native county sting nominal 0 0 O 2 d o io 15 class number nominal 0 e ee ee ee 16Kxindex _integer_ _ continuous_ 1 C Automatica _ Add Filter in Data Set 44 Cancel CUSTOMER SAP Infinitelnsight 7 0 213 2014 SAP AG or an SAP affiliate company All rights reserved _ Infinitelnsight Modeler Segmentation Clustering Defining a Variable Structure There are three ways to define a variable structure by first extracting the categories from the variable statistics then editing or validating the suggested structure by importing the structure from an existing model by building a new structure from scratch The option Optimal Grouping allows you to let Data Encoding group together the categories groups defined in the variable structure if they bring the same information For more details on variable structure see Infinitelnsight Modeler Regression Classification gt Defining a Variable Structure Filtering the Data Set In order to accelerate the learn process and to optimize the resulting model you can apply a filter to your data Set For this scenario Do not use the filtering option vV To Filter a Data Set 1 Check the option Add a Filter in Data Set 2 Click Next 12 capital loss number continuous O B a a Ss Sa eC Tae __ Lipatve couninisring _frominal_ _ _p _ __ __ _ _15class_ number nominal 0 Ce ee S vV To Add a Co
293. s bubbles The coordinates of a given bubble are the cluster centroid values according to two selectable continuous variables The size of the bubble is plotted according to the frequency of the corresponding cluster C calendar table A calendar table is used to ease the development of solutions around any business model which involves dates Acommon practice Is to have a calendar table pre populated with some or all of the needed information enabling to accomplish most date related complex tasks with simple database queries category A category is one of the possible values of a discrete variable A discrete variable is a nominal or ordinal variable It is the basic element used to code the variable as well as to gather descriptive statistics CUSTOMER SAP Infinitelnsight 7 0 273 2014 SAP AG or an SAP affiliate company All rights reserved Glossary category significance The category significance measures the impact acategory has on the target centroid Imaginary point inside a polygon whose coordinates are generally those of the polygon center chunk by chunk Number of lines of a table that are processed as package classification rate ratio between the number of correctly classified records and the total number of records confidence The Confidence of a rule is a measure that indicates the percentage of sessions verifying the consequent among those verifying the antecedent For instance the number of sessions con
294. s for the fitting check the box Use Score Bin Frequency as Weights 5 2 2 Step 2 Generating and Validating the Model Once the modeling parameters are defined you can generate the model Then you must validate its performance using the predictive power KI and the prediction confidence KR f the model is sufficiently powerful you can analyze the responses that it provides in relation to your business issue see Step 3 Analyzing and Understanding the Model Generated see page 113 and then apply it to new data sets see Step 4 Using the Model see page 154 Otherwise you can modify the modeling parameters in such a way that they are better suited to your data set and your business issue and then generate new more powerful models CUSTOMER SAP Infinitelnsight 7 0 109 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Generating the Model vV To Generate the Model 1 On the screen Specific Parameters of the Model click the Generate button The screen Training the Model appears The model is being generated A progress bar allows you to follow the process if SAP InfiniteInsight Vx y z class_Census01 File Help Training the Model a E pda amp Stop View Type Copy Print Save Export to ba PowerPoint Starting Model learning Stop Current Task 44 Cancel 4 Previous I gt Next 2 Ifthe Autosave option has been activ
295. s for which you want to add the probabilities in the output file CUSTOMER SAP Infinitelnsight 7 0 171 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Miscellaneous Outputs Outlier Indicator This option allows you to show in the output file which observations are outliers An observation is considered an outlier if the difference between its predicted value and its real value exceeds the value of the error bar In other words the error bar is a deviation measure of the values around the predicted score It appears in the output file as outlier_rr_ lt target variable gt Possible values are 1 if the observation is an outlier with respect to the current target else O Predicted Value Quantile This option allows you to cut the output file in quantiles and to assign to each observation the number of the quantile containing it Approximate quantiles are constructed based on the sorted distribution and the boundaries of predicted scores from the validation sample The score boundaries are used to determine approximate quantiles on the apply data set 1 Note Exact quantile computation would require a full sort of the scores obtained on the apply data set which can be resource consuming SAP Infinitelnsight V6 0 offers Gain Chart option for this purpose It appears in the output file as quantile_rr_ lt target variable gt _ lt number of quantiles gt for exa
296. s selected 3 Click Next The screen Summary of the Modeling Parameters appears 4 Goto the section Checking Modeling Parameters Checking Modeling Parameters The screen Summary of Modeling Parameters allows you to check the modeling parameters just before generating the model 4 SAP InfiniteInsight VX y 2 cass_Census01 File Help Summary of Modeling Parameters Model Name class Census01 Description Kxen SmartSegmenter Data to be Modeled _ _ Samples Census Census01 csv Cutting Strategy Random without test Target Variable class Weight Variable Optional None Find the best number of clusters in this range fio s fio Calculate SQL Expressions W KEN Export KxShell Script 44 Cancel 41 Previous I Generate l Note The screen Summary of Modeling Parameters contains an Advanced button By clicking this button you access the screen Specific Parameters of the Model For more information about these parameters Setting Up the Advanced Options on page 223 CUSTOMER SAP Infinitelnsight 7 0 222 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering The Model Name is filled automatically It corresponds to the name of the target variable class for this scenario followed by the underscore sign _ and the name of the data source minus its file extension Census01 for this scenario Before generating the model you
297. screen Summary of Modeling Parameters appears 3 Go back to the section Check Modeling Parameters Validating the Model Once the model has been generated you must verify its validity by examining the performance indicators The predictive power allows you to evaluate the explanatory power of the model that is its capacity to explain the target variable when applied to the training data set A perfect model would possess a predictive power equal to 1 and a completely random model would possess a predictive power equal to O The prediction confidence defines the degree of robustness of the model that is its capacity to achieve the same explanatory power when applied to a new data set In other words the degree of robustness corresponds to the predictive power of the model applied to an application data set To see how the predictive power and the prediction confidence are calculated see Predictive Power Prediction Confidence and Model Graphs on page 40 Beside the Predictive Power KI and the Prediction Confidence KR SAP Infinitelnsight also provides you with two commonly known indicators the Classification rate in case of a classification model the Pearson Square Correlation coefficient named R2 in SAP Infinitelnsight in case of a regression model Both indicators can be used to compare SAP Infinitelnsight results with results obtained through other data mining tools l Note Validation of the mode
298. scription file desc_census csv These files allow you to evaluate SAP Infinitelnsight features and take your first steps in using it CensusOl csv is the sample data file that you will use to follow the scenarios of Infinitelnsight Modeler Regression Classification and Infinitelnsight Modeler Segmentation Clustering This file is an excerpt from the American Census Bureau database completed in 1994 by Barry Becker l Note For more information about the American Census Bureau see http www census gov CUSTOMER SAP Infinitelnsight 7 0 201 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering This file presents the data on 48 842 individual Americans of at least 17 years of age Each individual is characterized by 15 data items These data or variables are described in the following table Variable age workclass fniwgt education education nu m marital status occupation relationship race SEX capital gain capital loss native country class 1 Note Description Age of individuals Employer category of individuals Weight variable allowing each individual to represent a certain percentage of the population Level of study represented by a schooling level or by the title of the degree earned Number of years of study represented by a numerical value Marital status Job classification Position in family Ethn
299. se fe SAP InfiniteInsight Vx y z File Help Select a Data Source a Use a File or a Database Table Use Explorer Data Type Text Files Metadata are stored in the same place as data source Folder U U Samples Census 44 Cancel EJ You must select a data set 41 Previous I Next CUSTOMER SAP Infinitelnsight 7 0 67 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 2 Click the Browse button The following selection dialog box opens Select Source Folder for Data oF fi gd BG C Usersinatacha yam Documents Ay A HE Census H JapaneseData H El KelData H E KSN E E KTS l Ci i i i i g JA KA E PA Tl Ti i Tl i ld Samples ext Files dat data csv tt Use r Passwo rd oK Cancel 3 Double click the Samples folder then the Census folder l Note Depending on your environment the Samples folder may or may not appear directly at the root of the list of folders If you selected the default settings during the installation process you will find the Samples folder located in C Program Files SAP Infinitelnsight InfinitelnsightVx y z 1 Select the file CensusOl csv then click OK The name of the file appears in the Estimation field 2 Click Next The screen Data Description appears a SAP InfiniteInsight Vx y Z New Regression Classification Model Fi
300. served Infinitelnsight Modeler Regression Classification Understanding the Classification Decision Screen The screen Classification Decision allows you to either select a percentage of the population who will respond positively to your campaign of Detected Target or a percentage of the entire population of Population When moving the cursor on the scale the different values are updated accordingly For example if you select the option of Detected Target and set the cursor to 80 the value of the field of Population will be 32 0 which means that if you want that 80 of the people who will respond positively to your campaign receive your mailing you will have to send it to 32 of the entire population On the other hand if you select the option of Population and set the cursor to 20 on the scale the value of the field of Detected Target will be 60 4 which means that if your budget only allows you to send your mailing to 20 of the entire population you will touch 60 of the population who will respond positively For more details on how to use the Confusion Matrix see section Analyzing and Understanding the Model gt Confusion Matrix on page 144 CUSTOMER SAP Infinitelnsight 7 0 163 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Using the Option Direct Apply in the Database Pre requisites for the In database Apply Mode This optimize
301. sification Advanced Apply Settings General Outputs Copy the Weignt Variable This option allows you to add to the output file the weight variable if it had been set during the variable selection of the model Copy Data Set Id This option allows you to add to the output file the name of the sub data set the record comes from Estimation Validation or Test A Caution This option cannot be used with the in database apply feature Copy tne Variables This option allows you to add to the output file one or more variables from the data set vV To Add All the Variables Check the All option vV To Select only Specific Variables 1 Check the ndividual option 2 Click the gt gt button to display the variable selection table 3 Inthe Available list select the variables you want to add use the Ctrl key to select more than one variable 4 Click the gt button to add the selected variables to the Selected list CUSTOMER SAP Infinitelnsight 7 0 165 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification User Defined Constant Outputs This option allows you to add to the output file constants such as the apply date the data set name or any other information useful for using the output file A user defined constant is made of the following information Parameter Description Visibility indicates if the constant will appear in the output or not Name
302. sight Modeler Segmentation Clustering Axes The X axis shows the influence of the variable categories on the target The significance of the different numbers on the X axis are detailed in the following table Number on the X axis Indicates that the category has positive number a positive influence on the target O no influence on the target the behavior is the same as the average behavior of the whole population negative number a negative influence on the target The Y axis displays the variable categories Categories sharing the same effect on the target variable are grouped They appear as follow Category_a Category_b Category_c Categories not containing sufficient numbers to provide robust information are grouped inthe KxOther category When a variable is associated with too many missing values the missing values are grouped in the KxMissing category Both categories are created automatically by SAP Infinitelnsight Formulas Category Importance NP BF NC where VPis the Normal Profit BF is the Bin Frequency and NC is the Normalization Constant The calculation of the normalization constant differs by target data type The calculations for binary and continuous targets are detailed below For binary targets it is calculated as follow Target Frequency 1 Target Frequency It can be approximated for non pathological continuous targets that is continuous targets without distribution peak Dirac from
303. ss 4 4 Building 50L pass 1 Building 50L pass Building 50L pass 3 Building 50L pass 4 Building 50L pass 5 Preparing the engine 1 1 for statistics computation on Validation Computing statistics Preparing the engine 1 1 for statistics computation on Estimation Computing statistics Covers overlap 0 361293 Frequency of unassigned records 0 005350497 Indicator on validation Quality KI 0 740313 nnn AA Robustness KR 0 99299 Learning time 20 seconds Smart Segmenter learn finished Total elapsed time 7 seconda End of the training process 44 Cancel You can then display the screen Using the Model a If the performance of the model satisfies you go to Step 3 Analyzing and Understanding the Model Generated on page 232 b Otherwise go to the procedure To Generate a New Model To Generate a New Model Either click Previous to return to the modeling parameters defined initially Then you can modify the parameters one by one Or click Cancel to return to the main screen of Modeling Assistant Then you must redefine all the modeling parameters CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 6 2 3 Step 3 Analyzing and Understanding the Model Generated IN THIS CHAPTER USEE NOI acta se encase rte ya E ooze ye iercesdes lease eco ne seiendense uekeoen die aeetceuetoene fee
304. ss of the model that is its capacity to achieve the same explanatory power when applied to a new data set In other words the degree of robustness corresponds to the predictive power of the model when applied to an application data set To discover how these indicators are calculated Predictive Power Prediction Confidence and Model Graphs on page 40 l Note Validation of the model is a critically important phase in the overall process of Data Mining Always be sure to assign significant importance to the values obtained for the predictive power and the prediction confidence of a model CUSTOMER SAP Infinitelnsight 7 0 37 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 7 Under what Circumstances is a Model Acceptable Quality Indicator the Predictive Power No minimum threshold is required for the predictive power of a model This depends upon the context of your work that is your domain of application the nature of your data and your business issue In some cases a model with a predictive power as low as 0 1 may allow realization of a profit of several thousands dollars In all cases a positive predictive power indicates that the model generated will perform better than a random model Robustness Indicator the Prediction Confidence A model with a prediction confidence inferior to 0 95 must be considered with caution The performance of such a model is very likely to vary between the
305. st possible prediction of the value of the target variable for each observation of the data set Random red The profit that may be achieved using 25 of the initial data set using a random model 25 belonging to curve at the a random model that does not allow the target category of the target variable are selected bottom one to know even a single value of the target variable for each observation of the data set Predictive Power Prediction Confidence and Model Graphs On the model graph plot Ofthe estimation data set default plot the predictive power corresponds to the area found between the curve of the model generated and that of the random model divided by the area found between the curve of the perfect model and that of the random model As the curve of the generated model approaches the curve of the perfect model the value of the predictive power approaches 1 Ofthe estimation validation and test data sets select the corresponding option from the list Data set located below the plot the prediction confidence corresponds to one minus the area found between the curve of the estimation data set and that of the validation data set divided by the area found between the curve of the perfect model and that of the random model CUSTOMER SAP Infinitelnsight 7 0 239 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Category Significance Definit
306. t vV To Display the Exact Profit Values for a Given Point On the screen Model Graphs on the plot click a point on one of the curves presented vl To Select the Debriefing Type 1 Onthe screen Model Graphs above the plot click the drop down list associated with the Debriefing Type field The list of debriefing types appears Debriefing Type Predicted ws Actual Predicted vs Actual Actual vs Predicted h 2 Select a debriefing type The corresponding plot appears CUSTOMER SAP Infinitelnsight 7 0 121 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Understanding Model Graphs For a Model with a Nominal Target The following figure represents the model graph produced using the default parameters Profit Type Detected v Models rr_class v Performance 0 9 4 0 8 0 7 4 0 6 0 5 4 Detected Profit 0 4 4 0 3 4 oh ph wor ar aor ane ap ape or oP er ee aor ep 40 ar or oe op op ee La La percentage E Random E Wizard Validation CUSTOMER SAP Infinitelnsight 7 0 122 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification On the plot the curves for each type of model represent the profit that may be realized Y axis that is the percentage of observations that belong to the target variable in relation to the number of observations sele
307. t contains a form to fill which reproduces the Infinitel nsight model ScoreCard only available for Infinitel nsight Modeler Regression Classification models l Note When generating SQL and SAS codes you will be asked to provide the names of the key column and of the data set used CUSTOMER SAP Infinitelnsight 7 0 188 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification A Caution Only SQLServer key code handles trimmed data during its execution For other codes if data are not trimmed it may generate some differences Advanced Settings UNICODE Mode The option Activate UNICODE Mode allows you to generate the code selected in Unicode so that it supports non latin languages such as Japanese Russian and so on l Note This option is particularly useful for SQL codes SOLZUDF Options The option Do not generate code for non contributive variables allows you to exclude from the code all variables with a contribution of O since they do not influence the result In some cases this can significantly reduce the size of the generated code You can either Use the default separator GO or Use a custom separator CUSTOMER SAP Infinitelnsight 7 0 189 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Exporting the Model as a KxShell Script The KxShell script export allows you to gen
308. t means that the relation between the variables and the target variable has changed as a consequence the model should be rebuilt on the new data If the KI and KR are not much different it means that the relation between the input variables and the target behavior has not changed but it does not mean that differences of distributions are not possible Control for Deviations Reports The panel Control for Deviations provides you with six options that can be separated in three groups the first one made of the options Probability of Deviation Probability of Category Deviation and Probability of Grouped Category Deviation enumerates the probabilities of deviation of each variable distribution be it by variable variable category or group of categories A probability over 0 95 indicates that the variable or category global distribution is significantly different in the control data set than in the reference data set l Note The probability of deviation is actually a standardized Chi test It is significant above 0 95 CUSTOMER SAP Infinitelnsight 7 0 157 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification the second group comprised of the options Probability of Target Deviation and Probability of Target Deviation for Grouped Categories lists for each variable the probabilities of deviation of the categories and the grouped categories with respect to the target variab
309. t model would possess a predictive power equal to 1 and a completely random model would possess a predictive power equal to O The prediction confidence defines the degree of robustness of the model that is its capacity to achieve the same explanatory power when applied to a new data set In other words the degree of robustness corresponds to the predictive power of the model applied to an application data set For more details on the model results see section Model Summary For this Scenario The model generated possesses A quality indicator KI equal to 0 808 A robustness indicator KR equal to 0 992 The model performs sufficiently well You do not need to generate another To Validate the Model Generated 1 Verify the Predictive Power KI and Prediction Confidence KR of the model These indicators are marked in red on the following figures g SAP InfiniteInsight Vx y z class_Census01 File Help Model Overview gt a Copy Print Save Export to PowerPoint Report Type Model Overview hai w Y CI V O WY Model class Census01 DataSet Census01 csv Initial Number of Variables 16 Number of Selected Variables 14 Number of Records 48 842 Building Date 2014 05 06 09 25 07 Learning Time 19s Engine Name Kxen RobustRegression Author natacha yam Nominal Targets TargetKey 1 0 Frequency 76 05 1 Frequency 23 95 Selection Process Selected Iteration 1 Predictive Power KI 0 809 Predictio
310. t the variable to be used to expand the next level of the decision tree a Automatically expand the next level using the most contributive variable not yet used in the current decision tree a Fold the section of the tree displayed below the current node The thickness of the arrows depends on the amount of population in the node In the following example the arrow leading to the node corresponding to the category 0 4386 of capital gain is thicker since the node population is significantly higher than the one from the node capital gain 4386 41310 Node Details When you select a node the node information is displayed in the tab Node Details located in the lower part of the panel bee Ui Profit Curve Detected FS Target class Selected Sub population Whole Population Estimation Validation 36361 12461 arig 2973 CUSTOMER SAP Infinitelnsight 7 0 149 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification This tab indicates the target for which the current decision tree is displayed and provides you with the following information for each data set in the model Population Count that is the number of records found in the current node For continuous targets Target Mean that is the mean of the target for the current node For nominal targets Positive Target Count that is the number of records for which the target is positive Positive T
311. table 41 Previous I OK CUSTOMER SAP Infinitelnsight 7 0 97 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification 3 Set the parameters listed in the following table Parameter Model Name Description Data Type Folder File Table 4 Click OK 98 CUSTOMER Description This field allows you to associate a name with the model This name will then appear in the list of models to be offered when you open an existing model This field allows you to enter the information you want such as the name of the training data set used the polynomial degree or the performance indicators obtained This information could be useful to you later for identifying your model Note that this description will be used instead of the one entered in the panel Summary of Modeling Parameters this list allows you to select the type of storage in which you want to save your model The following options are available Text files to save the model in a text file Database to save the model in a database Flat Memory to save the model in the active memory SAS Files to save the model in a SAS compatible file for a specified version of SAS and a specified platform SAS v6 or 7 8 for Windows or UNIX SAS Transport to save the model in a generic SAS compatible file Depending upon which option you selected this field allows you to specify the ODBC source t
312. taining the item D among the ones containing the itemset A B C confusion matrix The confusion matrix allows visualizing the target values predicted by the model compared with the real values and setting the score above which the observations will be considered as positive that is the observations for which the target value is the one wanted consequent Y is called the consequent of the rule The consequent is composed of only one item for example Y can be the item D CUSTOMER SAP Infinitelnsight 7 0 274 2014 SAP AG or an SAP affiliate company All rights reserved Glossary continuous variable Continuous variables are variables whose values are numerical continuous and sortable Arithmetic operations may be performed on these values such as determination of their sum or their mean contribution relative importance of each variable in the built model correlation Any measure that quantifies the fact that two variables share the same information This can be measured by looking at the relative variation of the two variables for different entities Classical statistics defines linear correlation to compute such a metrics on continuous variables SAP Infinitelnsight can compute correlations between variables of different types by looking at the correlation of the codes of both variables in presence of a target cross Statistics A method of estimating the accuracy of a classification or regression model The data set
313. tandard errors on validation L1 0 302877 Index of current iteration 6 Reference EI 0 808648 Reference KER 0 995696 Number of kept variables 3 KI 0 746901 ER 0 992623 L2 0 L2 0 For the current iteration 3 variables kept KI 0 Computing statistics Computing statistics Indicator on validation Quality EKI 0 808648 Robustness KR 0 995696 Standard errors on estimation L1 0 266717 Standard errors on validation L1 0 268897 Learning time 18 seconds L2 0 L2 0 Classification regression learning phase finished Total elapsed time 15 seconds End of the training process a oll CUSTOMER 360722 362271 334565 335906 Click the 1a JE Stop Current Task button Click the Previous button The screen Summary of Modeling Parameters appears Go back to the section Check Modeling Parameters Progression button The progression bar screen appears Lint 0 990842 Lint 0 937751 T46901 KR 0 992623 Linf 1 24811 Linf 1 06301 co Om SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Validating the Model Once the model has been generated you must verify its validity by examining the performance indicators The predictive power allows you to evaluate the explanatory power of the model that is its capacity to explain the target variable when applied to the training data set A perfec
314. target A large KI difference has been observed for this variable between Estimation and Validation data sets with respect to the target It will be excluded from the model with respect to this target The variable has a small KR with respect to the target It will be excluded from the model with respect to this target z This option allows you to display the current report view in the graphical table that can be sorted by 2 View This option allows you to display the current report view as a HTML table values or by ascending or descending alphabetical order You can also select which data should be displayed Some reports can be displayed as a pie chart 1 Some reports can be displayed as a bar chart This bar chart can be sorted by ascending or descending 7 Some reports can be displayed as a line chart i J BA When the current report is displayed as a bar chart this option allows you to change the orientation of zZ Sort the bars from horizontal to vertical and vice versa This option allows you to display the current report with no sorting tt This option allows you to sort the current report by ascending values N This option allows you to sort the current report by descending values 3 This option allows you to sort the current report by ascending names 3 This option allows you to sort the current report by descending names CUSTOMER SAP Infinitelnsight 7 0 140 2014 SAP AG or an SAP affiliate comp
315. tatus Remove You can add or remove variables from this list and view the model variables structure as explained below 9 Once all the variables for which you want to import the structure from the model are displayed in the list Click OK The selection window closes and the structure state changes vV To Adda Variable to the List of Variables 1 Inthe list Variable from Loaded Model select the variable you want to add to the list of variables for which the structure will be imported 2 Click the Add button X K2R_Census_331_1 x Target from Loaded Model m Variables from Training Census0 1 csv Variables from Loaded Model K2R_Census_331_1 a Ge Avew View Censusi Lcsyv K2R_Census_331 Version 1 marital status OK Cancel The variable appears in the list below CUSTOMER SAP Infinitelnsight 7 0 79 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification vV To Remove a Variable from the List of Variables 1 Inthe list located in the lower part of the panel select the variable for which you do not want to import the structure 2 Click the Remove button X K2R_ Census 331 1 j x Target from Loaded Model dass ka Variables from Training Census01 csv workclass ka Variables from Loaded Model K2R_Census_331_1 workclass ka dh Add a View K2R_Census_331 Version 1 marital status The variable is re
316. te gov T516 Bachelors 13 Never marr Adm clerical Not in far 2 50 Self emp n 63311 Bachelors 13 Married civ Exec mana Husband 3 36 Private 215646 H5S grad 9 Divorced Handlers c Mot in far 4 53Frivate 234721 11th T Married civ Handlers c Husband 5 26 Private 336409 Bachelors 13 Married civ Prof specia Wife 6 37 Frivate 204582 Masters 14 Married civ Exec mana Wife T 49 Private 160187 9th 5 Married sp Other seri Not in fan 8 Self emp n 209642 H3S grad 9 Married civ Exec mana Husband g 31 Private 45781 Masters 14 Never marr Prof specia Not in fan 10 42 Private 159449 Bachelors 13 Married civ Exec mana Husband 11 37 Private 280464 Some colle 10 Married civ Exec mana Husband 1 30 State gov 141297 Bachelors 13 Married civ Prof specia Husband 13 23 Private 122272 Bachelors 13 Never marr Adm clerical Own child 14 32 Private 205019 Assoc acdm 12 Never marr Sales Not in fan 15 40 Private 121772 Assocvoc 11 Married civ Craft repair Husband 16 34 Private 245487 7th 8th 4 Married civ Transport Husband 17 25 Selfempen 176756 HS grad 9 Never marr Farming fis Qwn child 18 32 Private 186624 H3S grad 9 Never marr Machine o Unmarrie 19 38 Private 28687 11th T Married civ Sales Husband 20 43 Self emp n 292175 Masters 14 Divorced Exec mana Unmarrie 21 40 Private 193524 Doctorate 16 Married civ Prof specia Husband Close 2 Inth
317. ted automatically by SAP Infinitelnsight Category Importance Definition The following definition applies to continuous targets some wording may be simplified for binary targets The formulas presented below can also be applied to the binary target case use categories instead of segments in this case We consider the case where a Infinitelnsight Modeler Regression Classification regression model is trained on a continuous target signal S with the help of an input variable X Infinitelnsight Modeler Regression Classification starts by binning the continuous target S into B segments peeey the target We will suppose that the input X is a nominal categorical variable though the whole process can be extended easily to the case of ordinal and continuous inputs We will Suppose that X has N categories X Xy We are interested in assessing the importance of a category Xi with respect to the target S The importance of a category depends on two factors The fact that the distribution of the target for this category is significantly skewed towards high values or low values when compared with the distribution of the target on the entire population The frequency of this category High importance can result from either of the following a high discrepancy between the target distribution for cases associated to this category and the distribution of the target variable for the entire population aminor discrepancy
318. test phase of your campaign whose responses to the campaign are known This sample thus constitutes a training data set This sample taken from the complete database also exhibits some missing values Your business issue thus consists of Rapidly building a segmentation model using the training data set or sample The clusters obtained will allow you to better understand the profiles of the individuals in your database as a function of their propensity to purchase Then applying the segmentation model obtained from the training data to the entire list of prospects to determine which cluster each individual should belong CUSTOMER SAP Infinitelnsight 7 0 198 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 6 1 5 Your Solutions To select the individuals to whom you should send a mailing there are several solutions you can use An intuitive method A classical statistical method K means both ascending and descending hierarchical segmentation models The Infinitelnsight method Intuitive Method This method consists of using your knowledge of the various profiles exhibited by your customers Thanks to the domain specific knowledge that you have of your customers you determine the criteria of the segmentation model intuitively and build the clusters yourself The main disadvantage of this method is that the number of information items available for
319. the data set you can generate a second model by excluding the variables too closely correlated with the target variable For this Scenario Exclude the variable Kx ndex as this is a key variable Since the initial data set does not contain a key variable SAP Infinitelnsight feature generated KxIndex automatically Retain all the other variables To Exclude Variables from Data Analysis 1 Onthe screen Selecting Variables in the section Explanatory Variables Selected left hand side select the variable to be excluded A SAP InfiniteInsight VX Y 1 New Regression Classification Model File Help Selecting Variables Explanatory Variables Selected 14 Target Variables 1 workclass fnlwat education education num marital status occupation relationship race sex _ Alphabetic Sort capital gain Weight Variable O capital loss hours per week psa Excluded Variables 1 Ey C Aphatetic sort L Alphabetic Sort CUSTOMER SAP Infinitelnsight 7 0 95 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification l Note On the screen Selecting Variables variables are presented in the same order as that in which they appear in the table of data To sort them alphabetically select the option Alphabetic sort presented beneath each of the variables list 2 Click the button gt located on the left of the screen section Excluded Variables lower right han
320. the viewing options that interest you For more information about viewing options Definition Depending on the type of the target the model graph plot allows you to View the realizable profit that pertains to your business issue using the model generated when the target is nominal Compare the performance of the model generated with that of a random type model and that of a hypothetical perfect model when the target is nominal Compare the predicted value to the actual value when the target is continuous On the plot for each type of model the curves represent 118 when the target is nominal the realizable profit on the Y axis as a function of the ratio of the observations correctly selected as targets relative to the entire initial data set on the X axis when the target is continuous the predicted value or score on the X axis in respect with the actual value or target on the Y axis CUSTOMER SAP Infinitelnsight 7 0 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Plot Options vV To Display the Graphs for the Estimation Validation and Test Sub sets Click Data Set and select one of the following options that allow you to switch between the graph for the Validation sub set LY the graphs for all the sub sets M l vV To Change the View Type Click View Type and select the desired option vV To Copy the Model Graph Click the A
321. to 1 500 of your customers You know which product A or B each customer has purchased 10 000 prospects You want to know which product each customer is likely to purchase The variable product purchased is your target variable it corresponds to your business issue It Is Known for all values of the training data set in our example the customers Not known for the values of the application data set in our example the prospects SAP Infinitelnsight features allow you to model that target variable and thus predict which product each of your prospects is likely to purchase The following table represents your database Name Age Residence Socio Occupational Category Product Purchased Charles 34 New Orleans Manager Administrator Product A John 37 Washington Manager Administrator Product A Marl ne 31 Boston Civil servant Product B Prospect 1 34 Oakland Manager Administrator Prospect 2 24 Washington Civil servant Prospect n 35 Sacramento Skilled tradesman Constraints Governing Use The following constraints govern the use of a target variable Within a training data set all target variable values must be known Only binary or continuous variables may be used as target variables Explanatory Variable Definition An explanatory variable is a variable that describes your data and which serves to explain a target variable CUSTOMER SAP Infinitelnsight 7 0 32 2014 SAP AG or an SAP affiliate compa
322. to technical constraints a data set corresponding to the database of 1 000 000 customers that will be used in this scenario cannot be provided to you CUSTOMER SAP Infinitelnsight 7 0 173 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification You will apply the model to the file CensusOl csv which you used to generate the model In this manner you will be able to compare the predictions provided by the model to the real values of the target variable Class for each of the observations In the procedure To Apply the Model to a New Data Set M i In the section Application Data Set select the format of the data source Select the format Text files In the Generate field select the option Individual Contributions Select the folder of your choice in which to save the results file Model Generated Output Do not select the option Save only outlier observations To Apply the Model to a New Data Set On the screen Using the Model click the option Apply model The screen Applying the Model appears a SAP InfiniteInsight VX y 2 class_Census01 File Help Application Data Set Data Type Tex Files ha W Browse a i Browse Define Mapping Generation Options Generate Predicted Value Only Advanced Apply Settings Mode Apply ha Results Generated by the Model Data Type Tex Files w Folder J J Samples Census
323. tory variables in relation to the target variable This significance is relative as the weight of each variable is pro rated as a function of the significance of the other explanatory variables 4 SAP InfiniteInsight VX y 2 cass_Census01 File Help aly Contributions by Variables Bar Orientation asa S Ta rt to Copy Print Save EXPO Excel Pin View Chart Type Maximum Smart Variable Contributions wt Maximum Smart Variable Contributions 0 000 0 025 0 050 0 075 0 100 0 125 0 150 0 175 0 200 0 225 0 250 marital status capital gain occupation educatian numi age capital loss hours perweek education Variables relationship Sex wo tke lass native country race tnilvagt The plot above corresponds to the model generated and illustrates the two variables that contribute the most to the target variable which in this scenario are marital status capital gain In other words the marital status and capital gain variables are those which have the greatest effect on whether a prospect will respond positively or negatively to your marketing campaign Among all the variables included in the sample data set these two are the most discriminatory variables with respect to the target variable Class CUSTOMER SAP Infinitelnsight 7 0 127 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Corre
324. training data set and the application data sets 4 7 8 How to Obtain a Better Model Obtaining a better model is achieved by Improving the prediction confidence of the model or Improving the predictive power of the model or _ Improving both the predictive power and he prediction confidence of the model Several techniques allow you to improve these indicators You can increase the degree of complexity of the model polynomial degree The following table presents other techniques To improve You can The predictive power of a Add variables to the training data set model Use combinations of explanatory variables that seem relevant to you The prediction confidence of a Add observations to the training data set model l Note For more information about improving the predictive power and the prediction confidence Indicators Specific to SAP Infinitelnsight see page 39 CUSTOMER SAP Infinitelnsight 7 0 38 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts 4 8 Performance Indicators 4 8 1Indicators Specific to SAP Infinitelnsight Two indicators allow you to evaluate the performance of a SAP Infinitelnsight model The predictive power which is the quality indicator of the model The prediction confidence which is the robustness indicator Quality Indicator Predictive Power Definition The predictive power of a model is the quality indicator of models g
325. ts the expected deviation of the current model The blue area shows where about 70 of the actual values are expected to be In other words it means that in case of a Gaussian distribution about 70 of the actual points should be in the blue area keep in mind that this is a theoretical percentage that may not be observed every time The default setting for the type of curve parameter is Predicted vs Actual The extreme values for prediction ranges are TargetMean sqrt TargetVariance TargetMean sqrt TargetVariance 1 Note sqrt TargetVariance iS equal to the Standard Deviation CUSTOMER SAP Infinitelnsight 7 0 124 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification Predictive Power Prediction Confidence and Model Graphs On the model graph plot Of the estimation data set default plot the predictive power corresponds to the area found between the curve of the model generated and that of the random model divided by the area found between the curve of the perfect model and that of the random model As the curve of the generated model approaches the curve of the perfect model the value of the predictive power approaches 1 Of the estimation validation and test data sets select the corresponding option from the list Data set located below the plot the prediction confidence corresponds to one minus the area found between the curve of the estim
326. uble click the bar of the variable which interests you In case no user structure has been defined for a continuous variable the plot category significance displays the categories created automatically using the band count parameter The number of categories displayed corresponds to the value of the band count parameter For more information about configuring this parameter please refer to the section Band Count for Continuous Variables Plot Options VI To Switch Between Validation Data Set and All Data Sets Plots 1 Click Data Sets and select the 279 All Data Sets button to display all data sets The plot displaying all data sets appears a SAP InfiniteInsight VX y 2 cass_Census01 File Help ali Category Significance D B Copy Print Save EXPO tO pin View z Excel Data Sets View Type _ Bar e r Orientation Variables age ka Variable age Influence on Target 0 075 0 050 0 025 0 000 0 025 0 050 0 075 0 100 0 125 0 150 0 175 0 200 categories Ta ais ha 2 23 24 19 22 17 19 E Estimation Validation 44 Cancel CUSTOMER SAP Infinitelnsight 7 0 241 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering 2 Click Data Sets and select the 2 Validation Only button to go back to the Validation Data Set plot V To Switch between Curve and Bar Charts 1 Click View Type and select the aj butt
327. ue O is assigned to observations that do not belong to the target category of the target variable The value 1 frequency of the target variable in the data set is assigned to observations that do belong to the target category of the target variable The following table describes the three curves represented on the plot created using the default parameters The curve Represents For instance by selecting Wizard green The profit that may be achieved using 25 of the observations from your entire data set with the help of a curve at the top the hypothetical perfect model perfect model 100 of observations belonging to the target that allows one to know with absolute category of the target variable are selected Thus maximum profit is confidence the value of the target achieved variable for each observation of the data set 1 Note These 25 correspond to the proportion of prospects who responded in a positive manner to your marketing campaign during your test phase For these prospects the value of the target variable or profit is equal to 1 Validation The profit that may be achieved using 25 of the observations from your initial data set with the help of blue curve in the the model generated by the model generated 66 9 of the observations belonging to the middle I nfinitel nsight Modeler target category of the target variable are selected Regression Classification that allows one to perform the be
328. upon the format of the results file generated use Microsoft Excel or another application to open the file The figure below presents the headings and columns of the results file obtained for this scenario KxIndex class ke_clusterld kc_TargetMeanClustld 1 0 3 0 017524 2 0 5 0476401001 3 0 3 0 017524 4 0 2 0237075999 B 5 0 5 0476401001 G 0 4 0308998985 EJ 7 0 3 0 017524 9 a 1 5 0476401001 9 1 1 0 942696989 10 0 942696959 11 1 5 0476401001 12 1 5 0476401001 13 0 3 0 017524 14 0 4 0308898985 15 1 2 0237075999 16 0 2 0237075999 2 You can now analyze the results obtained and use these results of your analysis to make the right decisions CUSTOMER SAP Infinitelnsight 7 0 269 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Description of the Results File Depending upon which options you selected the results file will contain some or all of the following information in the same order as seen below The key variable defined during data description at the model parameter settings step If your data set did not contain a key variable the key variable Kx ndex would have been generated automatically by SAP Infinitelnsight Possibly the target variable given as known values if the latter appeared in the application data set as is the case in this scenario The variable kc_clusterld which indicates the number of the cluster to which each observati
329. ure Group Structure i Category Edition he ASSOC VOC J Bachelors i Bachelors J Masters Masters di Doctorate Prof school i Doctorate i Prof school Jo HS grad i HS grad i 50ome college i Some college Ji 1st 4th Preschool KxMissing Ada New Group Add Category he Istath J Lt Preschool me om KxMissing New Category Remove Group Remove Category Merge Add 2 OK Cancel CUSTOMER SAP Infinitelnsight 7 0 85 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification A To Create a New Category In the field right of the button New Category enter the name of the category to add Click the button New Category The category is created in the list Category Edition To Add Categories to a Group In the list Category Edition select the category or categories to add to a group In the list Group Structure select the group in which you want to add the selected categories Click the button Add Category To Delete a Group In the list Group Structure select the group to delete Click the button Remove Group All the categories belonging to this group are re added to the list Category Edition To Remove a Category from a Group In the list Group Structure select the category or categories you want to remove from the group Click the button Remove Category The selected categories are removed from
330. ure of deviation from uniform response rates across categories of a variable Kolmogorov Smirnov is a non parametric exact goodness of fit statistic based on the maximum deviation between the cumulative and empirical distribution functions In the case of a binary classification task people are interested by the difference between the Lorenz curve see page 48 for the good cases 1 a and the Lorenz curve for the bad cases B when selecting an increasing ratio of population These curves evolve from O to 1 together and the K S statistics is the maximum deviation between these two curves For a perfect system the K S statistics is 1 and that for a random system because of the equality between the two curves the K S statistics is O l Tip The K S is used to calculate the difference between two distributions in order to have an idea about the quality of a data set AUC The AUC statistic is a rank based measure of model performance or predictive power calculated as the area under the Receiver Operating Characteristic curve see ROC on page 47 For a simple scoring model with a binary target this represents the observed probability of a signal responder observation having a higher score than a non signal non responder observation For individual variables ordering based on score is replaced by ordering based on the response probability for the variable s categories for example cluster ID or age range response rates The co
331. use queries that are considered too heavy In any case you should check with them to know the line of action to follow CUSTOMER SAP Infinitelnsight 7 0 69 2014 SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Regression Classification The Explain Mode has not been Configured If your DBMS Administrator has not configured the Explain mode the following pop up opens when you try to access the data X Query Validation F No longer request validation for similar queries Show Details You need to contact your Administrator who will tell you which action to take and configure the Explain mode If the Administrator validates the execution of the query you may want all queries with the same duration to be executed without validation In that case check the box Do not request validation anymore for similar requests The validation message will then only appear for larger queries This configuration will only be used for the current session when closing SAP Infinitelnsight it will be lost For a permanent configuration see your DBMS Administrator who will find the necessary information in the support document Explain Mode available in section Support and Integration Documentation of SAP Infinitelnsight documentation Describing the Data Selected For this scenario Select Text Files as the file type Use the file Desc_CensusOl csv as the description file for th
332. use they were not measured not answered were unknown or were lost monotonicity The direction of variation of a monotonic function does not change multiple instance installation An installation mode that consists in running several instances on one server in order to divide up the load N nominal variable Nominal variables are variables whose values are discrete that is belong to categories and are not sortable Nominal variables may be Numerical meaning that its values are numbers Textual meaning that its values are character strings Important binary variables are considered nominal variables normalize To transform numerical data and make them fir into a defined interval CUSTOMER SAP Infinitelnsight 7 0 283 2014 SAP AG or an SAP affiliate company All rights reserved Glossary numeric filter A digital filter is a system that performs mathematical operations on a sampling discrete time signal to reduce or enhance certain aspects of that signal O ordinal variable Ordinal variables are variables with discrete values that is they belong to categories and they are sortable Ordinal variables may be numerical meaning that its values are numbers They are therefore ordered according to the natural number system O 1 2 and so on textual meaning that its values are character strings They are therefore ordered according to alphabetic conventions outlier A data value that do
333. value The odds ratio is equal to 1 p p p is the probability of risk Performance 450 400 350 4 300 4 a 250 pej pe o 200 150 100 50 0 n an n NO ah A N pes P gr E Ed P o gt gO gr G2 GP go a e 40 score Validation CUSTOMER SAP Infinitelnsight 7 0 51 2014 SAP AG or an SAP affiliate company All rights reserved Essential Concepts Probability of Risk The X axis represents the risk score and the Y axis represents the odds ratio value The probability of risk p is computed for each risk score bin this way number of Bad divided by the number of records in the risk score bin Performance Probability a o a o Lee ais ii ari Sa ha o ts ane pa ah ape pe ee ah ef ta ah g5 gi ae a an ai ate age gi at 40 Score E validation Population Density The density is computed according to the number of records in each risk score bin 20 by default Performance 1 600 1 500 1 400 1 300 1 200 i 1 100 1 000 00 a00 700 600 s00 300 200 100 he E Density score E Validation CUSTOMER 52 2014 SAP AG or an SAP affiliate company All rights reserved SAP Infinitelnsight 7 0 Essential Concepts Risk All All three curves are displayed in the same graph Note that the y axis of the probability curve is on the right hand side The y axis of the population density and the good bad odds on the left Risk P
334. variable that the explanatory variables are able to explain R random cutting strategy The random cutting strategy distributes the data of the initial data set in a random manner between the three sub sets estimation validation and test record The fundamental data structure used for performing data analysis Also called a table row or example A typical record would be the structure that contains all relevant information pertinent to one particular customer or account robustness The degree of robustness corresponds to the predictive power of the model applied to an application data set robustness indicator the prediction confidence KR The prediction confidence is the robustness indicator of the models generated using SAP Infinitelnsight It indicates the capacity of the model to achieve the same performance when it is applied to a new data set exhibiting the same characteristics as the training data set CUSTOMER SAP Infinitelnsight 7 0 286 2014 SAP AG or an SAP affiliate company All rights reserved Glossary ROC The ROC Receiver Operating Characteristic graph is derived from signal detection theory It portrays how well a model discriminates in terms of the tradeoff between sensitivity and specificity or in effect between correct and mistaken detection as the detection threshold is varied role In data modeling variables see page 292 may have three roles They may be Target variables see page 290
335. ven a cluster A taken from this data set the same variable gender may be distributed as follows 80 of observations belong to the category male 20 of observations belong to the category female This distribution corresponds to the profile of the variable gender over cluster A The clusters profiles allow you to view and compare the profiles of the variable gender over the data set and the clusters taken from this data set CUSTOMER SAP Infinitelnsight 7 0 252 2014SAP AG or an SAP affiliate company All rights reserved Infinitelnsight Modeler Segmentation Clustering Displaying Clusters Profiles vV To Display Cluster Profiles 1 On the screen Using the Model click Clusters Profiles The screen Clusters Profiles appears a SAP InfiniteInsight VX y 2 cass_Census01 File Help ali Cluster Profiles E A W XLS a amp 2 aSG View Type Bar sort Reset Cluster Copy Print Save Export to Orientation Categories Names Excel rr Frequencies of T a 25 24 E i i Cluster Name 2 3 715 BB 5 88 4 244 E 2 23 5 E 227 o O 6 E 55 es 25 51 7 E iro 4 aow EEE Variables occupation ka E Fix Variable Cluster 1 vs Whole Population for Variable occupation E b 1 486 Lo 1 Categories E All Population Cluster 1 2 Inthe table select the cluster for which you want to view the profile l Note If only
336. ws calculation of the profit that may be realized using the model In general a benefit is associated with the positive or expected values of the target variable and a cost is associated with the negative or unexpected values For instance in the context of a promotional mailing campaign a person is associated with A benefit for responding to the promotional mailing A cost for not responding to the promotional mailing 4 9 2 Available Profit Types To visualize the profit that may be realized using a model generated by the SAP Infinitelnsight you may use the following profit types Detected profit Lift profit Standardized profit Customized profit Detected Profit Detected profit is the profit type shown as the default It allows examination of the percentage of observations belonging to the target category of the target variable that is the least frequent category as a function of the proportion of observations selected from the entire data set Using this profit The value O is assigned to observations that do not belong to the target category of the target variable The value 1 frequency of the target category of the target variable in the data set is assigned to observations that do belong to the target Lift Profit Lift profit allows examination of the difference between a perfect model and a random model and between the model generated by SAP Infinitelnsight and a random model It re
337. y of the target variable than the mean l Note You can display the profit curve for the selected variable by clicking the button we Display Profit Curve located in the tool bar under the title The importance of a category depends on both its difference to the target category mean and the number of represented cases High importance can result from a high discrepancy between the category and the mean of the target category of the target variable or a minor discrepancy combined with a large number of records in the category or acombination of both The width of the bar shows the profit from that category The positive bars correspond to categories which have more than the mean number from the target category that is responders and the negative bars correspond to categories which have less than the mean number from the target category that is responders The Variables pull down menu allows the selection and graphing of any of the variables in the model The tool bar located under the title allows the user to copy the coordinates to the clipboard print the plot or save it in PNG format The values are normalized and their sum always equals to O Depending on the chosen profit strategy or on the continuous target variables value type you can obtain all positive importances or negative and positive importances CUSTOMER SAP Infinitelnsight 7 0 244 2014 SAP AG or an SAP affiliate company All rights reserved Infiniteln
338. y these models CUSTOMER SAP Infinitelnsight 7 0 290 2014SAP AG or an SAP affiliate company All rights reserved Glossary time series A time series is a sequence of data points measured typically at successive times spaced at often uniform time intervals timeout A specified period of time that will be allowed to elapse before a specified event is to take place unless another specified event occurs first time stamped population A time stamped population is a list of pairs lt identifiers time stamps gt the semantic meaning of such a construct can be associated with snapshots of the entities and a given time in general terms a given entity may be represented at different time stamps in a single time stamped population training Another term for estimating a model s parameters based on the data set at hand training data set A training data set is a data set used for generating a model By analyzing the training dataset SAP Infinitelnsight features will generate a model that allows explanation of the target variable based on the explanatory variables transaction A transaction is defined by a unique key the key of the related session an attribute called an item CUSTOMER SAP Infinitelnsight 7 0 291 2014 SAP AG or an SAP affiliate company All rights reserved Glossary true negative correct assignments to the class of non signals true positive correctly identified signal U
339. yed in the column kc_ lt TargetVariable gt _Mean the difference with the actual target value if the latter is Known for the current observation displayed in the column kc_ lt TargetVariable gt _Error for nominal targets the proportion of the least frequent category of the target variable key category in the cluster containing the current observation displayed in the column kc_ lt argetVariable gt _Mean Types of Results Available The application of a model to a data set allows you to obtain three types of results The cluster index for each observation The disjunctive encoding or dummy coding of the cluster indexes which means that for each cluster a boolean variable is created indicating whether the current observation belongs to that cluster or not For a given observation the value 1 is assigned to the variable corresponding to the cluster containing the observation and the value O is assigned to the variables corresponding to the other clusters The variable names are built according to the following pattern kc_ lt TargetName gt _ lt ClusterIndex gt Consider as an example that you have generated a five clusters model When applying this model SAP Infinitelnsight creates five variables corresponding to the five generated clusters For an observation belonging to cluster 3 the result appears as shown below KxIndex class kc_class kc class 1 kc class 2 kc_class_3 kc_class_ 4 kc class 5 I gt

Classification, Regression, Segmentation and Clustering

Contents

Download Pdf Manuals

Related Search

Related Contents