SAS數(shù)據(jù)挖掘?qū)嵗?廣告投放問題eng教學(xué)文稿

1、Good is good, but better carries it.精益求精，善益求善。SAS數(shù)據(jù)挖掘?qū)嵗?廣告投放問題eng-SASTutorial:EnterpriseMinerCasedescriptionSupposethatyouworkforamailorderenterprisethatsendsoutacatalogoffurnishingsandhousewareseachmonth.Aspartofanupcomingsalescampaign,youwanttodistributeaspecialcatalogthatisdevotedtofinediningandc

2、ontainskitchenware,dishes,andflatware.Itstooexpensivetosendthiscatalogtoallofyourcustomers,soyouneedtotargetthosemostlikelytobuy.Youdothisbydevelopingatargetingmodelandthenusingittoproduceanewmailinglist.Youhaveanextensiverecordofcustomerpurchases.Thedataincludesvariablesthatindicatewhethercustomers

3、boughtkitchenware,dishes,orflatwareinthepasttwoyears.ThispurchasehistoryhasbeenusedtocreatetheCUSTDET1dataset,whichcontains49variableswiththefollowinglabels:PurchaseDollarsSpentYearlyIncomeHomeValueOrderFrequencyRecencyMarriedNamePrefixAgeSexTelemarketInd.RentsApartmentOccupied0intheDINEBIN(N)=formu

4、lafieldatthebottomofthewindow.ClickOK.TheCreateVariablereappearswithdining0displayedintheFormulafield.ClickOK.ThenewvariableDINEBINappearsintheTransformVariableswindow.ClosetheTransformVariableswindow.Aconfirmationboxappears.SelectYestosaveyourchangesModifyingAttributesYounowneedtoidentifyitasthemod

5、elstarget.ThisisdonewiththeDataSetAttributesnodeDraganddroptheDataSetAttributesnodeontotheworkspacetotherightoftheinputdatasourcenode.ConnecttheTransformVariablenodetotheDataSetAttributesnode.Double-clicktheDataSetAttributesnode.TheDataSetAttributeswindowappears.ClicktheVariablestab.Scrolldownthelis

6、tofvariablesuntilDINEBINappears.Noticethegrayed-outcolumnModelRoleandthewhitecolumnNewModelRole.Grayed-outcolumnsreflecttheoriginaldatasetattributesandtheycannotbeedited.Rolereferstotheuseofeachvariable.Mostvariablesaretreatedasinputvariablesinanattempttopredictthetarget.Ifyouscrolldownthelistofvari

7、ables,youwillseethatEnterpriseMinerconsiderscertainvariablesunsuitableasinputs(e.g.,dates,orvariableswithasinglevalue).Suchvariablesaregiventherolerejected.Right-clickinthecolumnNewModelRoletotherightofthevariableDINEBIN.SelectSetNewModelRolefromthepop-upmenu.Selecttarget.Youaretryingtotargetthebuye

8、rsofdiningwares(forwhomthevariableDinebin=1).Howeverothervariablesinthedatasetcontainthesameinformation:KITCHEN,DISHES,andFLATWAREDINEBINhasvalueof1ifthecustomerhadboughtanydiningware.Itisthereforenecessarytoexcludethemfromtheanalysis(assigna“reject”status).Note:WithintheDataSetAttributesWindow,thec

9、olumnMeasurementreferstomeasurementlevel.Thisistherangeofvaluesthatisfoundineachvariable.Therearefivepossibleassignments:unary-onevalueforexample,avariablewithaparticularvaluethatwasusedtocreateadatasubsetbinary-twovaluesforexample,thevariableMARITALthatcontainsNoorYesnominal-morethantwonon-numericv

10、alues,butnoimpliedorderforexample,STATECODthatcontainsAK,AL,AR,AZ,etc.ordinal-morethantwobutnotmorethantennumericvalues,withimpliedorderforexample,NUMCARSthatcontainsvaluesfrom0to3interval-morethantennumericvaluesforexample,AMOUNTthatcontainsmanydifferentdollarvaluesAneworderforvaluesinTargetVariabl

11、eWhenyoubuildamodel,EnterpriseMinerconsidersthetargeteventtobethefirstsortedvalueofthetargetvariable.Thedefaultsortorderisascending.Butthenewtargetvariable,DINEBIN,containsvaluesof0and1,with1representingthepurchaseofanydiningwares.ThevaluesneedtobeindescendingorderforEnterpriseMinertoaimattheintende

12、dtarget.Tochangetheorderofatargetvariable:ClicktheClassVariablestabintheDataSetAttributeswindow.Scrolldownthelistofvariablesuntilthetargetvariableappears.Right-clickintheNewOrdercolumntotherightofthetaretvariable.SelectSetNewOrderfromthepop-upmenu.SelectDescending.AnewlevelInthecaseofthedatainDataMi

13、ning2003,thenewDINEBINvariablehasbeenassignedthewrongmeasurementlevel.ScrolldowntheDataSetAttributeswindowandnoticethattotherightofDINEBIN,tothecolumnMeasurement.Becausethevariablecontainsonlyvaluesof0and1,thecorrectmeasurementlevelisbinary.Right-clickintheNewMeasurementcolumntotherightofthevariable

14、DINEBIN.SelectSetNewMeasurementfromthepop-upmenu.Selectbinary.ThenewmeasurementlevelofDINEBINisreflectedinthewindow.DataPartitionYouhavecontroloverhowthepartitionsarecreated.TheDataPartitionnodeprovidesseveraloptionsDraganddroptheDataPartitionnodeontotheworkspace.ConnecttheDataSetAttributesnodetothe

15、DataPartitionnode.Double-clicktheDataPartitionnode.ItswindowappearswiththePartitiontabintheforegroundBuildaDecisionTreeModelDraganddroptheTreenodeontotheworkflow.ConnecttheDataPartitionnodetotheTreenode.Double-clicktheTreenodetoopenit.TheTreewindowappearswiththeVariablestabintheforeground.TheBasican

16、dAdvancedtabscontainasetofcriteriafortreebuildingandevaluation.Ifyouhavemadechangestothedefaultsettingsforthisnode,anewwindowpromptsyoutoenteramodelnameEnterDirect-TreeintheModelNamefield.EnterDirectMarketingDecisionTreeModelintheModelDescriptionfield.SelectOK.TheSaveModelAswindowclosesandreturnsyou

17、totheworkflow.ViewsoftreestructureRight-clicktheTreenode.SelectRunfromthepop-upmenu.Analertboxappears.SelectYestoviewyourresults.TheTreeResultswindowappearswiththeAlltabintheforeground.SelectthePlottabtoviewperformanceontrainingandvalidationsetsSelectViewTreetoviewtreestructureEvaluation/AssessmentT

18、heAssessmentnodetakesoutputfromanymodelingnodeandchecksthemodelsaccuracyagainstdatainthetestpartition.DraganddroptheAssessmentnodeontotheworkflowbelowtheTreenode.ConnecttheTreenodetotheAssessmentnode.Right-clicktheAssessmentnode.SelectRunfromthepop-upmenu.AstheDataMining2003dataisprocessedintheworkf

19、low,eachnodeinturndisplaysathickgreenborder.Whenitdoes,youarepromptedtoviewtheresults.SelectYes.TheAssessmentToolwindowappearsNote:TheAssessmentnodedoesnotanalyzeallavailabledata,itreliesonsamplingtoproducequickresults.Asaconsequence,differentliftchartsshowvariationsevenwhentheyusethesamedataLiftcha

20、rtsTheModelstaboftheAssessmentToolwindowlistsmodelsthatyouhavedefined.Scrollingtotherightdisplaysstatisticsforeachmodel.SelecttheDirect-Treemodel.SelectToolsLiftChartfromtheSASmenubar.TheLiftChartwindowappears.Thebluebaselinerepresentstheresponseratethatyouobtainbynotusingamodel,butbysendingthecatal

21、ogtoeveryoneinyourcustomerdatabase.DiagnosisChart(ConfusionMatrix)SelectToolsDiagnosischartReportresultsTheReporternodeistheEnterpriseMinertoolfordrawingadetailedmapoftheDataMiningprocess.DraganddroptheReporternodetotheworkspace.ConnecttheAssessmentnodetotheReporternode.Right-clicktheReporternode.Se

22、lectRunfromthepop-upmenu.SelectOpenwhenthereportgenerationmessageboxappears.AnextensivereportonyourminingworkflowappearsinyourdefaultWebbrowser.Applyingthemodelonadataset(Trainingorother)TheScore!nodeperformsthistask.DraganddroptheScore!nodetotheworkspace.Connectthetreenodetothescorenode.Draganddrop

23、theInputDataSourcenodeontotheworkspaceabovetheScore!node.ConnecttheInputDataSourcenodetotheScore!node.Double-clicktheInputDataSourcenode.TheInputDataSourcewindowappears.Asyoudidatthebeginningoftheworkflow,specifyadatasetastheSourceData.Fornow,selecttheEMDATAdirectory,andselectthefilethatbeginswithts

24、t(plussomerandomcharacters)tobethesourcedata.Thisisthetrainingpartition.ChangetheRoleofthedatasetfromRAWtoSCORE.ClosetheInputDataSourcewindow.Aconfirmationboxappears.SelectYestosaveyourchanges.GaininsightDouble-clicktheScore!nodethatisalreadyintheworkflow.TheScore!windowappears.Selecttheradiobuttont

25、hatisnexttoApplytrainingdatascorecodetoscoredataset.ClosetheScore!window.Aconfirmationboxappears.SelectYestosaveyourchanges.DraganddroptheInsightnodeontotheworkspacebelowtheScore!node.ConnecttheScore!nodetotheInsightnodeDouble-clicktheInsightnode.TheInsightSettingswindowappears.ClicktheSelectbuttont

26、hatisnexttotheDatasetfield.TheImportsMapwindowappears.Youneedtoclickthroughthehierarchicallistofdatasetstoidentifythedatasetthatisassociatedwiththescoredata.ThisdatasettypicallyhasanSDprefix,followedbyastringofrandomalphanumericcharacters.FindthefileClicktheplussign(+)totheleftofScore!Applyinthedata

27、setlist.ItexpandstoshowSAS_DATA_SETS.Clicktheplussign(+)totheleftofSAS_DATA_SETS.Itexpandstoshowfourdatasets.SelectthedatasetwhosefilerefbeginswiththelettersSD.TheRoleandDescriptionfieldsidentifyitasscoredata.ClicktheOKbutton.YoureturntotheDatatab.IntheInsightbasedonfield,clicktheradiobuttonthatisnexttoEntiredataset.ClosetheI