User Manual () - Decision Support Sciences
Contents
1. e If the MiningSolve M server is open and connected through a network to the computer with the MiningSolve M Root Server the Root Server will automatically locate the server and begin generating models with it e Availability of the server can be customized by selecting Advanced from the Availability menu on the LE server In the top portion of the dialog a specific time can be specified to allow a connection To set up a schedule of availability use the Custom option and the bottom section of the dialog will become available e Either the times to allow a connection or the times to block a connection are shown Select the Allow Connection or Don t Allow Connection options to change which options are viewed Select one or more days and a range of times then select the Add button to add an additional set of criteria for the server availability MiningSolve x 210 x T HiningSoly Server File Availability Statistical Server Help Current Connection Computer JON Start Time 02 22 2002 09 06 4M Runs Completed ee Average Run Time 00 00 16 Run Status i Receiving Data File Receiving Run Specificatons gt Syntax Generation iL Transfering Commands to Modeling Server Generating Model amp Determining Model Performance Sending Run Results to Client Connection Progress Number of Connections 3 Elapsed Time 00 23 31 Estimated Remaining Time 01 02 31 Rune Completed 187 of 20012. MiningSolve Training Manual Using the Decision Support Sciences Data Mining Automation Engine To Automate Effective Marketing Strategy Development Decision Support Sciences Better Science Better Solutions e O Introduction What is Data Mining MiningSolve M uses many statistical techniques to optimize the predictions of specified variables MiningSolve runs thousands of runs automatically in order to find the highest performing models MiningSolve M can produce and scan thousands of models in the time 1t takes to put together one model with a conventional statistical package Rather than manually setting up each analysis run the user can efficiently evaluate only the solutions MiningSolve M determines are the best solutions Using the technology of distributed computing MiningSolve M can be run on many computers at once connected by a local area network LAN or the internet This leverages existing computing power to run thousands of models per day efficiently and economically MiningSol ve 2 Decision Support Sciences Table of Contents I Analysis Setup 1 Using Rules Files 2 Using the Rules Wizard II Data Mining III Appendix Dialog Boxes MiningSol ve 3 Decision Support Sciences I Analysis Setup 1 Using Rules Files 2 Using the Rules Wizard MiningSol ve 4 Decision Support Sciences Analysis Setup Using Rules Files Using Rules Files E Rules Files e A rules file st
3. Modeling Server Priority te arme Current Run Algorithm Method LogReg BStep Cond Criteria 0 09 Transformation 0 8461539461 53646 EE Hamal Above Normal Disconnect HiningSoly Server X When to Allow Connection C Always C Never 4 56 PM C In 0 hours 0 minutes Custom Custom Connection Schedule Da Stattimee Stop Time CS afe Weekdays 12 00 AM 8 00 AM de Weekdays 6 30 PM 11 59 PM Weekends All Day All Day T Sunday M Monday Tuesday M Wednesday I Thursday IV Friday ST Saturday Weekends Between 30AM and fe I NextDay Allow Connect All Day C Don t Allow Connection Add Remove All Decision Support Sciences Data Mining Execute Execute Data Mining E Execution Summary e When a run has been completed the MiningSolve M Execution Summary will be d e l d Run Server Algorithm Method Criteria Transformation isplayed 0 KRISTI 93 81 46 7 Multiple Discriminant Analysis Rao s V 003 1 32 30 JON 93 81 46 7 Multiple Discriminant Analysis Rao s 0 03 1 34 o s 27 KRISTI 93 75 46 67 Multiple Discriminant Analysis Mahalanobis Distance 0 09 0 98 2 This information 1S sorted m descending order of 8 KRISTI 93 65 4661 Multiple Discriminant Analysis Mahalanobis Distance 0 04 1 36 e e 41 JON 93 65 4661 Multiple Discriminant Analysis Mahalanobis Dist 0 09 0 66 Lift which puts the best runs at the top 24 KRIS
4. Decision Support Sciences Reference Manual dialog boxes 4 Rules Wizard Step 3 Build Mirrored Database In this step the database that will be used for the analysis is selected and imported To open a database select the Add button An Open File dialog will appear Select a file and click on the Open button and it will be imported into MiningSolve M If a database name was already selected for the current rules file but has the wrong path or file name use the Remove button to delete the name and reselect it e To select the variables to be used from a database highlight the database name on the wizard screen The available variables will appear in the Source Fields box Use the mouse to highlight any number of fields from the Source Fields box then click on the arrow to move the fields to the Mirrored fields box The mirrored fields are those that will be included in the data mining execution If as in this example N and I fields are being used there should be at least two I and one N mirrored fields e To select fields from a different database highlight the database name and select the fields to mirror e Select the Next box to move on to the next screen MiningSolve 27 HMiningSoly Wizard Step 3 Build Mirrored Database What data would vou like to give MiningSoly access to in order to a solve Your query Selected databases files FileName T9 C Program Files Market Advantage Mi
5. Multiple Regression n Forced Entry Forward Entry n Backward Elimination i Stepwise Selection Back Cancel Decision Support Sciences Reference Manual dialog boxes 7 Rules Wizard Step 6 Select Cases for Inclusion The data file can be filtered based on segment variables Multiple segments can be selected so that for example only people with a loan account who have been with the bank for more than 10 years are included in the run Missing data is data that is either system missing or marked as missing in SPSS Select one of the options to change how missing data is handled by MiningSolve M Missing data can be handled differently for discrete or continuous variables If segments have been selected the Working File Size reflects the number of cases included in the selected segments Use the slider to change the selected number of cases to use from the working data file and for the holdback sample MiningSolve x HinngSols Wizard Step 6 Select Cases for Inclusion af You can further clarity the model Wining olw will use by selecting specific respondents by either region or segment Mep Select respondents by region Missing Data Replacement Methodology Discrete Data Eliminate Cases Replace With Median For Two Group Variables Only Replace Using Multiple Discriminant Analysis Continuous Data Eliminate Cases Replace with Mean Replace Using Multiple Regression Replace Usi
6. generate 1707 syntax scripts Advanced W Cutoff at ed HiningSoly Wizard Step 7 Advanced Parameter Settings Do you want additional control over the way MiningSoly will work with the data in oh order to further optimize results Hand selecting the method order or increasing data transformation granularity may improve overall performance and speed gpoeseesecesoepeeesesooosoosoessesogesoogoesseosoesocosesseossesossoesooog Method Select how methods should be included in the analysis when at least one method is to be left out n Order Below Random Simultaneous Entry Wilks Lambda Mahalanobis Distance Smallest F Ratio Unexplained Variance Transformations Minimum Exponent 0 5 Maximum Exponent 1 5 Transformation Selection Nondinear Linear Move Down MiningSolve Cancel E Define Parameters e This dialog box displays the analysis specifications for each data mining method The parameters can be set the same for all algorithms or uniquely for each algorithm Use the radio buttons under the Data mining method box to specify your choice If Use the same settings for all algorithms is selected All algorithms must be selected under Data mining method in order to move the sliders To customize each algorithm individually set the parameters for one method then select the next method to customize Use the mouse to move the sliders to change t
7. three options for handling this data The first option leaves the incompatible data field out of the particular method for which it is inappropriate The second option attempts to use all of the selected data fields Data that is not compatible with a specific method will be recoded i e a continuous variable will be transformed using exponents to be used as a discrete variable The third option eliminates the analysis method rather than the data field if there is incompatible data Use the mouse to select the option by clicking on it then select the Next button MiningSolve 29 HiningS oly Wizard Step 5 Recommended Approaches pf Please select from the following list the proper data mining methodology to use in this query i Use the maximum number of methods eliminating fields not appropriate to each specific method f Use the maximum number of methods recoding fields not appropriate to each specific method Use only appropriate methods deleting from the query all methods requesting unmatching types The appropriate algorithme for your selection include E Multiple Discriminant Analysis BH Logistic F egress ioni i i Forced Entry z Backward Stepwise Conditional Statistic Backward Stepwise Likelihood Ratio z Backward Stepwise Wald Statistic z Forward Stepwise Conditional Statistic Forward Stepwise Likelhood Ratio i Forward Stepwise wald Statistic
8. 25 Decision Support Sciences Reference Manual dialog boxes 3 Rules Wizard Step 2 Define Goal e This dialog box allows you to select the desired business problem and displays the Please select the approach that best describes the problem you are mje tying to solve or the question vou are trying to answer appropriate algorithms for that problem dl i e Use the mouse to select the desired business ma the most Helene customers based Luren value b AI th 1 b f h d t Find the most profitable customers based lifetime value TODIEM oritnms WI e reiresne O h ld b d fe Identify the best prospects based on a cross sell show on y those that wou e used to Identify the best prospects based on any predictor determine solutions Identify the best customers to sell to within time constraints Cc Determine customer loss or non retention When Jou que satisfied with VIAL selection C Determine customer loss within time constraints click on the Next button C Determine the current product life cycle e If the Cancel button is selected The appropriate algorithme for your selection include MiningSolve M will exit this dialog box and EH cancel any changes that were made Smallest F A atig i Unexplained Varnance Logistic Regression l S Forced Entry i Backward Stepwise Conditional Statistic Backward Stepwise Likelihood Ratio Back Cancel MiningSolve 26
9. 4 08 5 KRISTI 93 47 46 51 Multiple Discriminant Analysis Smallest F Ratio 0 15 0 8 Percentage and the percentage of the holdback 37 JON 9347 4651 Multiple Discriminant Analysis Smallest F Ratio 017 1 22 6 KRISTI 93 46 46 5 Multiple Discriminant Analysis Smallest F Ratio 0 14 0 96 zi sample that was correctly classified e Lift is the correct classification percentage above chance alone This is determined by comparing the OCCP to the percentage achieved by chance alone e Transformation refers to the power that the data was taken to The range and steps for transformation can be set in the Define Parameters step of the Rules wizard by using the Advanced button e The specifications of each run are listed so that algorithms that performed well can be explored further for the current data set MiningSol ve 22 Decision Support Sciences lil Appendix Reference Manual MiningSolve 23 Decision Support Sciences Reference Manual dialog boxes 1 File Open Rules File e A rules file stores all of the files options and variables that have been selected while using MiningSolve M A rules file must be opened or created to begin the analysis e To Create a rules file select New Rules from the File menu before you start the analysis To use an existing rules file select Open Rules from the File menu e This dialog box will be opened that allows you to browse through all directories When the desired file has b
10. Approaches HMiningSoly Wizard Step 5 Recommended Approaches e The data variables that were selected for the analysis may not be appropriate for every analysis method that is used Please select from the following list the proper data mining orffae methodology to use in this query There are three options for dealing with data fields that Use the masimum number of methods eliminating fields not are incompatible with an analysis method RI AN p y Use the maximum number of methods recoding fields not appropriate to each specific method 7 Use the ee number of methods eliminating Use only appropriate methods deleting from the query all fields not appropriate to each specific method This EME EE ASE a option leaves the incompatible data field out of the Tissot care retto B Multiple Discriminate Analysis particular method for which it is inappropriate Direct Method Simultaneous i Wilk s Lambda Use the maximum number of methods recoding Mahalanobis Distance i Smallest F Ratio fields not appropriate to each specific method The cr second option attempts to use all of the selected data n E Logistic Regression fields Data that is not compatible with a specific E i i i Forward stepwise conditional statistic method will be recoded i e a continuous variable Forward stepwise likelihood ratio i be Forward stepwise Wald statistic will be transformed using exponents to be u
11. TI 9364 466 Multiple Discriminant Analysis RacsV 008 13 65 JON 93 64 46 6 Multiple Discriminant Analysis Rao s 0 08 1 4 33 KRIST 93 63 46 6 Multiple Discriminant Analysis Wilks Lambda 0 11 0 62 The Run column shows the order that the runs 13 JON 93 63 46 6 Multiple Discriminant Analysis Wilks Lambda 0 12 0 72 36 KRISTI 9361 46 58 Multiple Discriminant Analysis Wilks Lambd 0 03 O74 were completed The SPS and SPO files are saved 20 KPISTI 9359 4657 Multiple Discriminant Analysis Smallest F Ratio 014 14 e 59 JON 93 59 46 57 Logistic Regression Backward Stepwise Conditional Statistic 0 11 La for each run and named by the run number so if 1 KAISTI 9359 4657 Multiple Discriminant Analysis Wiks Lambda 018 082 Di A i 31 JON 93 59 46 57 Multiple Discriminant Analysis Mahalanobis Distance 0 05 0 52 you want to rerun a specific syntax file or inspect 2 KRISTI 9353 4654 MulipleDisciminentAnalsis Mahalanobis Distance 011 132 32 JON 93 53 46 54 Multiple Discriminant Analysis Mahalanobis Distance 0 12 1 26 the output later the run number will be necessary 42 KRISTI 93 5 46 52 Multiple Discriminant Analysis Smallest F Ratio EX F7 JON 935 46 52 Logistic Regression Backward Stepwise Conditional Statistic 0 05 0 5 a ye it KRISTI 93 49 46 52 Logistic Regression Forward Stepwise Conditional Statistic 0 1 0 6 OCCP stands for Overall Correct Classification 55 JON 93 49 4652 Logistic Regression Backward Stepwise Likelihood Ratio 00
12. clusion af Select Respondents by Geographical Region Im SNAN up Kokomo Indianapolis asti E i 11 Een Bay PA Manitowoc ieee i e sheboygan ay pa Ne Bay g g s sa Ra SA madison tGh skeadn rir NT Al at S Z LPDEPC nue E A S Ke nosh p Rockfo de Beloit LA SES x A Chicago ine A sal ar Fort Wayne r ee he And rson Ci Terre Haute Blo f O kansingt x MILPCHKA factor Ss MILPCHKB fl Ag 0 Zoomin M LPCDA E MILPCDB Lika ee Lie Zoom Out HHREGB Sais BE a VM HHREGC payini Z HHREGD Ban pa HREGE Park HHREGF JA Cincin Done MICUSTSEGI si Select All Matching Cases 971 48 ou can further clarify the model MiningS ol will use by selecting specific respondents by either region or segment Mau M Select respondents by market segment Segments Missing Data Replacement Methodology Discrete Data Eliminate Cases Replace With Median For Two Group Variables Only i Replace Using Multiple Discriminant Analysis Select respondents by region Continuous Data Eliminate Cases Replace with Mean Replace Using Multiple Regression Replace Using Maximum Likelihood Estimation Sample Size 2000 Valid Cases Working File Size 2000 100 Holdback Sample 758 recommended 1500 75 00 Defining market filters localizes your query onto respondents mje matching specif
13. e customer loss within time constraints Determine the current product life cycle The appropriate algorithms for your selection include Unexplained Varianace Rao s Y Logistic Regression Forced entry Forward stepwise conditional statistic Forward stepwise likelihood ratio Forward stepwise Wald statistic Backward stepwise conditional statistic Backward stepwise likelihood ratio Backward stepwise Wald statistic Multiple Regression x Cancel Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard E Build Mirrored Database e A database is required to provide the input for the MiningSoly Wizard Step 3 Build Mirrored Database data mining The input file must be an SPSS 7 5 or higher sav file SPSS should be installed on the ke beson e Tae cee er eee computer before using MiningSolve M Note SPSS should NOT be running when MiningSolve M is open if SPSS is running the MiningSolve M server will not connect to the root server Selected databases files Type Variables Cases C Program Files Market Advantage Minin SPSS 30 2000 e Select the Add button to open a browse dialog box to choose a file While the file is being loaded a progress indicator will tell you the database name o ni and fields that are being imported TA Eee Mirrored fields e Select the variables to use from the Source Fields box and m
14. een selected click the Open button e If the Cancel button is selected MiningSolve M will exit this dialog box without opening a rules file MiningSolve Open Look in EJ out a File name testrule rls Files of type DMAE Rules File Fr Cancel Decision Support Sciences Reference Manual dialog boxes 2 Rules Wizard Step 1 Introduction MiningSoly Wizard Step 1 Introduction x 1 Welcome to the MiningSoaly rules fle wizard This wizard will assist The rules file is the easiest and most effic F Way to step mae WOU in entering a set of rules for analysis in MiningSoly through all of the customizable options rerme data mining Start from scratch e To begin the rules wizard either select the icon from D lai thes bn tule Open a rules file to work from the toolbar or Start Rules Wizard from the Rules submenu of the Data Mining menu The rules wizard will present options on each screen that must be An SOSS Family Member determined to execute the data mining analysis Use the mouse to make your selection then click on the Next button Decision Support 3 Sciences e Atany point beyond the second step you can also go back to the previous rules wizard pages to modify your selections by using the Back button e If the Cancel button is selected MiningSolve M will exit the wizard and cancel any changes that were made Back Cancel MiningSol ve
15. eplace Using Maximum Likelihood Estimation file size slider Sample Size 2000 Valid Cases Working File Size 1825 91 25 ae e The Rules Wizard allows you to specify the holdback sample The holdback sample is the portion of the data that is excluded from the current analysis run Note the holdback eee css sample is NOT the sample of respondents being oe Li included in the analysis Holdback Sample FE recommended 1363 750 The percentage of the sample and number of respondents in the holdback sample is specified above the slider Back Cancel MiningSol ve 12 Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard HMiningSoly Wizard Step 7 Define Parameters af Please customize the ways in which you want MiningSaly to work With the data in order to get better output By selecting different algorithms from the drop down box vou can see what the values in the sliders mean for each particular approach Data mining method Multiple Discriminant Analysis Customize each algorithm individually TT Disable this i algorithm Use the same settings for all algorithms Methods 176 methods Criteria PIN varied from 0 03 to 0 18 EE Iterations Mot Applicable to this Algorithm ES Transformations powers of 0 50000 to 1 50000 by 0 02000 J Multiple Discriminant Analysis will generate 816 syntax scripts All algorithms together will
16. er to the root server each time the server is sending data to the root server When the server is receiving information a green ball is sent from the root server to the server The bottom dialog displays the Event Log screen This shows the status of completed runs and connection and analysis events on the servers and root server The I symbol on the left denotes informational stats the E symbol designates errors and the symbol signifies warnings The time that each server connects or disconnects is also shown Other screens with run information may be shown by selecting an option on the left side of the Runtime Information dialog MiningSolve 1 Runtime Information Center x Elapsed Time 00 00 18 Best Model Estimated Time Remaining 00 00 14 Algorithm Runs Completed 2 of 30 Method Servers Connected 3 Criteria Transformation 1 40 Progress E 7 OCCP 88 10 Multiple Discriminant Analysis Smallest F Ratio 0 03 Server Information Communications Performance s Best Run ie Event Log ER RR NX i Suspend All Running 30 of 3113 possible runs p mE Runtime Information Center x Elapsed Time 00 01 01 Best Model Estimated Time Remaining Done Algorithm Runs Completed 30 of 30 Method Servers Connected D Criteria 0 1 Transformation 0 60 Progress 100 OCCP 91 82 Server Information Communications Performance Mul
17. he level for that criteria If an algorithm is being customized individually the current level of the parameter is displayed on the right side of the box The combined effect of the parameter levels on the number of SPSS scripts to be generated is displayed below the last slider and changes dynamically as the level of any specific parameter is modified To further customize the analysis select the Advanced button at the bottom of the dialog to determine the order of the selected methods or the transformations settings For example if Rao sV is set as the first method in the advanced dialog AND In Order Below is selected and only 1 6 methods is selected on the Define Parameters dialog as the settings are shown then scripts will only be generated for MDA using the Rao s V method The total number of scripts to be generated is shown at the bottom of the dialog If a cutoff number is set MiningSolve M will randomly choose that number of scripts to run in the analysis 13 Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard E Define Output Depth and Holdback Sample MiningSoly Wizard Step 8 Define Ouput e In this dialog the user can select how MiningSolve M handles the output and which output is saved for future use e Changing the style of output affects how much information is included in the SPSS output files This does not affect the runtime statistics information MiningSol
18. ic characteristics By deselecting variables levels you can reduce total query time and get better results SA Pick Center TE La Navigate MW CONSLOAN W GENDER W LPDEPA MW LPDEPB Segment Levels Cases Searched 100 Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard E Select Cases for Inclusion cont Hining5oly Wizard Step 6 Select Cases for Inclusion e In data mining a sample of the population is analyzed and the results are evaluated for how You can further clarify the model MiningS als will use by selecting s ofae specific respondents by either region or segment accurately the population was characterized TT Select respondents by region klap The most reliable and real world method of iaia 4 Select dents by market t5 ts validation is to test the model on cases held back A a Segments from the analysis and compare what the model i E predicts to what we already know about the Eliminate Cases case This is called hold back sample validation C Replace With Median For Two Group Yariables Only C Replace Using Multiple Discriminant Analysis e The Sample Size reflects the number of valid Continuous Data cases taking into account any segments that Eliminate Cases 1 C Replace with Mean were selected A portion of the valid cases can nn C Replace Using Multiple Regression be randomly selected by adjusting the working R
19. ill also change the run that is displayed in the Best Model box at the top of the Runtime Information Center This change will be reflected on all dialogs of the Runtime Information Center MiningSolve 20 Runtime Information Center Elapsed Time 00 01 29 Estimated Time Remaining 00 01 22 Runs Completed 3 of 30 Servers Connected 1 Progress 10 E Server Information Communications Performance Best Model Algorithm Multiple Discriminant Analysis Method Smallest F Ratio Criteria Transformation 0 80 OCCP 96 94 of best run Jocce Algorithm Method Criteria Transformations Multiple Discriminant Analysis Smallest F Ratio 0 10 0 80 OCCP Lift L1 L2 A2 P2 m Performance Comparison 96 94 74 90 0 00 0 00 Additional Run Info Server Run Time Holdback size Model Size KRISTI 00 00 04 1505 495 Suspend C All Running 30 of 3113 possible runs fo Update Decision Support Sciences Data Mining Execute Execute Data Mining E Using the MiningSolve M Server e To generate models using MiningSolve there must be at least one server open and connected to the root server If MiningSolve M was installed to the default location select Start Program Files Decision Support Sciences MiningSolve M_Distributed MiningSolve M Server
20. ill appear in the lower text box e To select an algorithm use the mouse to click on it To open or close branches click on the or boxes on the left of the tree HiningSoly Wizard Step 2 Define Goal Please select the approach that best describes the problem you are oh trying to solve or the question you are trying to answer Find the most profitable customers based current value Find the most profitable customers based lifetime value Identify the best prospects based on a cross sell Identify the best customers to sell to within time constraints Determine customer loss or non retention Determine customer loss within time constraints Determine the current product life cycle The appropriate algorithms for pour selection include Multiple Discriminate Analysis Logistic Regression Multiple Regression Cancel MiningSolve one HiningSoly Wizard Step 2 Define Goal X Please select the approach that best describes the problem you are trying to solve or the question you are trying to answer Find the most profitable customers based current value Find the most profitable customers based lifetime value Identify the best prospects based on a cross sell Identify the best prospects based on any predictor Identify the best customers to sell to within time constraints Determine customer loss or non retention Determin
21. ion IF you select to have output files saved define the working directory for each of the selected the verbosity of their statistical reports methods and the syntax and output files are I Sore M Save output files verbose output saved in the subdirectories e The runtime statistics are saved in a CSV file i Ra ee in the working directory lt Back Cancel MiningSolve 32 Decision Support Sciences Reference Manual dialog boxes 10 Rules Wizard Step 9 Closure HiningSoly Wizard Step 9 Closure e The rules internal name is not a file name but a reference for the rules file that is used Congratulations you have finished defining a set of operating rules nn on ae for MiningSoly Please choose a name and an optional short within MiningSolve M and used for a run description of this set Then give this definition a directory into which nn temporary files should be written title The rules internal name and description are optional Rules Internal Name sa Sampl e The working directory specifies where all of iaia the output from the runs will be saved Description 2000 cases MDA runs predicting consumer loang Working Directory fe Program Files Market QdvantagesMiningSoly_ Distributed E Back Cancel MiningSol ve 33 Decision Support Sciences
22. iple Regression Runs Completed 63 of 500 En A Entry S ana Servers Connected 4 fiteria e A rules file must exist and be open to execute data mining li mi dre do RINASCE Tate 2 tL CURSE ODA Server information Computer Name Rune Completed Average Funtime stu 3 e From the MiningSolve Root Server select Execute from the Em ETRE IA Event Lo 00 enerating Mode Data Mining menu j Ra Sri di _ o n 00 00 00 Not connected hi e MiningSolve M servers will open SPSS and the sav data files oil p Current Model Best Model FN Active Servers that are linked to the open rules file and then use SPSS as a Algorithm MultReg Algorihm Mutfieg Cioe o Criteria 0 08 Criteria 0 16 V Never Connected Servers tool for executing the analysis Transformation 0 90 Transformation 050 53 i View Networ e The dialog at the right is the Server Information portion of the i a 92 53 60 64 Multiple Regression Forward Entr 0 16 Runtime Information Center All potential servers on the SOAS 5915 MuipleRegesion FomadEnty 003 7 i F u iple Iscriminan nalysis aos ae network are displayed along with their current status To see GSI daere MADE DrumwentAnapsis Meente Diane OTT T more information select one of the options on the left side of B883 4706 Multiple Regesson BackwadEimnaton O11 1 the dialog Suspend C All Running 500 of 1783 possible runs fo Update e The analysis can be paused at any time by using the Suspe
23. les submenu of Welcome to the MiningS aly rules file wizard This wizard will assist _ DAAE pou in entering a set of rules for analyste in Mining olw the Data Mining menu The first screen will allow you to 0 indicate whether you want to start from scratch modify an fora ni i Modit the existing rules existing file or open an existing rules file to work from Open a rules file to work from Start from scratch All settings will be blank or at the defaults Modify the existing Rules Settings from the open rules file will be pre set in the wizard and the modified rules file can be saved under a different filename at the end of the wizard Opena rules file to work from an Open File dialog box will appear when you click on the Next button Browse to select the file to use and click OK e The options to create a new rules file or open an existing one are also available manually from the File menu the New Rules and Open Rules menu items Back Cancel MiningSol ve 6 Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard E Define Goal e MiningSolve M is a powerful tool for determining optimum solutions to many different business problems In this dialog box the options for which problem to address and the techniques used to solve it are listed Use the mouse to select the problem to address When any problem is selected the algorithms that apply to that problem w
24. nd button To resume analysis select Restart Unfinished nee e IJ Execution from the Data Mining menu Looking In Search Complete Successful Entire Network af Networks Found 1 1 1 ci di 1 Et Microsoft Windows Network Workgroups Domains Found 1 e Displayed in the Best Model box at the top of the dialog are the ee ne specifications of the best model in the analysis thus far across all servers a e Select the Details button in the Best Model box for a prediction accuracy graph of each run The graph shows the prediction levels in numerical order not in the order in which they were performed gt DI dh Computers Near Me SERVER2 e Use the Add Server button to add a server that is not on the network i AARON si or that MiningSolve M does not find automatically To view computers on the local area network use the View Network button MiningSol ve 17 Decision Support Sciences Data Mining Execute Execute Data Mining E Execute cont There are several different views of information on a current MiningSolve M run The Communications and Event log dialogs are shown on the right On the communications dialog potential servers are shown as transparent images until they are connected When a server is active and connected to the root server the image will be darkened and the line connecting the server to the root server will turn blue On this visualization dialog a red ball is sent from the serv
25. ng Maximum Likelihood Estimation Sample Size 2000 Valid Cases Working File Size 2000 100 COO E Holdback Sample 75 recommended 1500 75 00 Decision Support Sciences Back Cancel Reference Manual dialog boxes 8 Rules Wizard Step 7 Define Parameters All of the methods and criteria used for the analysis can be customized using this dialog Select one of the Data Mining Methods to customize the options for that method Use the advanced button at the bottom to set specific exponent parameters or change the order of the methods to be used within each algorithm The number of SPSS scripts to be generated given the current settings is displayed at the bottom of the dialog A random selection of these possible scripts can be selected to be run by checking the Cutoff at box and entering a number of scripts to be run HiningSoly Wizard Step 7 Define Parameters HiningSoly Wizard Step 7 Advanced Parameter Settings Do you want additional control over the way MiningSoly will work with the data in one order to further optimize results Hand selecting the method order or increasing data transformation granularity may improve overall performance and speed purseeseezzazeoneeneesazesenecesezeonazeonenaezezaeneneanesizeenzenenne Method Select how methods should be included in the analysis when at least one method is to be left out In Order Below Random Simul
26. nng ol _ Distributed Sample sas SF Adad Remove Source fields Sample say Mirrored fields MM HHREGE Hew Feld Define Ippe Decision Support Sciences Back Cancel Reference Manual dialog boxes 5 Rules Wizard Step 4 Select Variables e Once the variables from all of the databases MiningS oly Wizard Step SSL have been selected the fields are divided into predictor and predicted fields Use the mouse to select the variables in the Mirrored fields box then use the arrows to Mirrored fields Field to predict move the fields to either the Fields to dg moe predict or the Fields to be used in one prediction boxes e Only N or O variables should be used as fields to predict Any variable type can be a predictor There should only be one field to predict and at least two predictor variables e When the desired fields have been selected click on the Next button if which data fields would you like MiningSol to schedule for DMAE prediction and which fields should be used to predict those variables Fields to be used in prediction MiningSol ve 28 Decision Support Sciences Reference Manual dialog boxes 6 Rules Wizard Step 5 Recommended Approaches e There may be some data fields that are not compatible with an algorithm that will be used in the analysis This step allows the user to choose how MiningSolve M will handle the incompatible fields There are
27. on Backward Elimination 0 07 44 12 Multiple Regression Stepwise Selection 0 02 45 78 Multiple Discriminant Analysis Mahalanobis Distance 0 02 Classification Percentages for all runs The Lift tab 44 06 Muliple Regression FowardEnty 001 Shows the lift which is OCCP above chance alone ei Chance is calculated by taking the sum of the 41 52 Multiple Discriminant Analysis Unexplained Variance 0 14 Na squares of the number of cases in each group divided by the square of the total number of cases Lift 100 OCCP 100 Chance e In the bottom half of the dialog the run statistics are shown These statistics can be sorted by any column by clicking on the column heading Use the scroll bar to view all of the statistics This list of statistics is also shown sorted by lift in the Execution Summary dialog when the runs have been completed Abort Suspend C All Running 750 of 1783 possible runs MiningSol ve 19 Decision Support Sciences Data Mining Execute Execute Data Mining E Execute cont The picture on the right shows the Best Run dialog from the Runtime Information Center This dialog displays detailed results for one run The best run that is shown may change based on the criteria used to select the best run To change the criteria used to select the best run use the mouse to click the arrow in the Criteria for best run box e Changing the criteria used to select the best run w
28. ores all of the analysis settings that you select for a run If the same database variables or algorithms will be used multiple times one rules file can be used as a template and modified each time e A rules file does not store the analysis reports or results but with an existing rules file the algorithms can be executed immediately E Open a Rules File e This is the first step when using MiningSolve M A new rules file may be created or an existing one can be used e To start the Rules Wizard Select Start Rules Wizard from the Rules submenu of the Data Mining menu as shown at the right E Using Rules Files e If an existing rules file is opened and you want to maintain the analysis preferences select Data Mining Execute e To modify the existing file or set the preferences for the new rules file use the Rules Wizard MiningSolve HiningSoly Root Server File Edit Data Mining STEP DELTA View Window Help Rogan e Uutput Gy Gant Gelert Vananles Start Rules Wizard Selec Sewe Define Boal Eustamze Alganithm Desta Eud Mimared Database Becammended Approaches Wanti Model Delne Parameters Define Outputs Aodiack Hame amp Wescibe Rules Erecute renew Bepart Bestar Uninehed Erection Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard E Using Rules Files x e To start the Rules Wizard select the LA button from the toolbar or Start Rules Wizard from the Ru
29. ove them to the Mirrored Fields box using the arrow button in the middle El Hew Beld Dette Ippe I NUMDEPD lt Back Cancel MiningSolve 8 Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard E Select Variables MiningSoly Wizard Step 4 Select Yariables e Select one variable to predict and at least one Which data fields would you like MiningSoly to schedule for variable to be used in prediction The field to ma prediction and which fields should be used to predict those predict must be a categorical variable with 2 or variables more non missing levels Highlight the variable s ete ARE in the Mirrored Fields box and use the arrows to MUTFUND N CONSLOAN l RETAINED move them to the box of selected variables e All of the available fields are listed in the Mirrored H fields list box categorized by type N designates a nominally encoded variable which means it can be stored as a binary number Any variable that can be asked as a yes no question would fit this category i e De Ae eset MI een gender have savings account etc This is a discrete variable I designates an interval scaled variable which when asked in an interview usually requires a text entry answer This is a continuous variable without level labels MiningSol ve 9 Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard E Identify Recommended
30. sed as a M erate rasa aaa Backward stepwise likelihood ratio i Rarkward steanwise MWiald statistic discrete variable Use only appropriate methods deleting from the query all methods requesting unmatching types This option eliminates the analysis method rather than the data field if there is incompatible data Back Cancel MiningSol ve 10 Decision Support Sciences Analysis Setup Rules Wizard Using the Rules Wizard E Select Cases for Inclusion e A specific segment of respondents can be extracted from the data for an analysis run These respondents can be selected either by region or by market segment The region option can only be used if location information has been included in the selected databases If the region option is checked a map dialog will appear Use the zoom and arrow controls to select the desired region Market segment Specific segments of the market can be included or excluded from the analysis Use the mouse to click on the checkmark to the left of the segment or level The levels for the highlighted segment appear in the Segment Levels box on the right Segments and levels with a checkmark will be included in the analysis Select the Done button when you are finished selecting segments The total number of cases to be used is displayed at the bottom of the dialog at Matching Cases MiningSolve HMiningSoly Wizard Step 6 Select Cases for In
31. taneous Entry Wilks Lambda Mahalanobis Distance Smallest F Ratio Transformations Minimum Exponent fos 0000 Maximum Exponent hs O Maximum Steps of CE Transformation Selection Nonlinear Linear Unexplained Variance Rao s Move Down MiningSolve 31 Please customize the ways in which You want MiningSaly to work with the data in order to get better output By selecting different algorithms from the drop down box you can see what the values in the sliders mean for each particular approach Data mining method Multiple Discriminant Analysis Customize each algorithm individually TT Disable this l algorithm Use the same settings for all algorithms Methods Direct plus 3 5 of the others SS Criteria PIM varied from 0 04 to 0 16 EE lherations Mot Applicable to this Algorithm Es Transformations powers of 0 60000 to 1 50000 by 0 12500 TE Multiple Discriminant Analysis will generate 360 syntax scripts All algonthms together will generate 362 syntax scripts W Cutoff at 1150 lt Back Next gt Cancel Decision Support Sciences Reference Manual dialog boxes 9 Rules Wizard Step 8 Define Output e Syntax and Output files that are saved are saved in the working directory specified in s i Please indicate which files you would like MiningSoaly to save for Step 9 of the wizard A folder IS created In mae further inspect
32. tiple Discriminant Analysis Smallest F Ratio i 9 16 544M on Tue Oct 09 2001 Run Completed Best Run i 9 16 554M on Tue Oct 09 2001 Run Started Si i 9 16 57AM on Tue Oct 09 2001 Run Completed i 9 16 584M on Tue Oct 09 2001 Run Started i 9 16 584M on Tue Oct 09 2001 Run Completed i 9 16 584M on Tue Oct 09 2001 Run Completed i 9 16 594M on Tue Oct 09 2001 Run Started i 9 16 594M on Tue Oct 09 2001 Run Started Y 9 17 004M on Tue Oct 09 2001 Run Completed 09 17 D1AM on Tue Oct 09 2001 Run Started i 9 17 03AM on Tue Oct 09 2001 Run Completed 9 17 034M on Tue Oct 09 2001 MininaSoly Execution Completed Details Decision Support Sciences Data Mining Execute Execute Data Mining Elapsed Time 00 21 52 m Best Model BI Execute C ont Estimated Time Remaining 00 22 57 Algorithm Multiple Regression Runs Completed 61 of 750 Method Stepwise Selection Servers Connected Criteria Transformation 1 10 Progress OCCP 92 32 The picture on the right shows the Performance dialog from the Runtime Information Center This dialog displays graphs and statistics for a set of runs Server Information To change the graph that is displayed click on one of the tabs below the graph The graphs are dynamically updated as the runs are completed RSR 0 5 36 43 Multiple Discriminant Analysis Rao s 05 0 The OCCP tab shows the Overall Correct 56 21 Multiple Regressi
33. ve i Analysis Setup Rules Wizard Using the Rules Wizard ia Closure HiningSoly Wizard Step 9 Closure e Itis best to name and save your settings as a rules file in aa i Congratulations ou have finished defining a set of operating rules case you want to use the same or similar settings in the mae for MiningSoly Please choose a name and an optional short description of this set Then give this definition a directory into which future temporary files should be written e The Rules Internal Name field is not a file name so does Piles intemal Name not have standard file name restrictions characters such Sample run as and are allowed DE on 2000 cases MDA runs predicting consumer loang e When the Finish button is selected you will be asked if you want to save the Rules file A Save As dialog box will appear and you can specify a filename and directory e When a rules file is complete select Execute from the Data VETTA orking Directory Mining menu to run the analysis cAProgram Files Market Advantage MiningS oly Distributed E Back Cancel MiningSolve 15 Decision Support Sciences II Data Mining MiningSol ve 16 Decision Support Sciences Data Mining Execute Execute Data Mining x Elapsed Time 00 01 32 Best Model Execute Estimated Time Remaining 00 05 36 Algorithm Mult
Download Pdf Manuals
Related Search