数据挖掘实例广告投放问题eng.docx

数据挖掘实例广告投放问题eng.docx

《数据挖掘实例广告投放问题eng.docx》由会员分享，可在线阅读，更多相关《数据挖掘实例广告投放问题eng.docx（12页珍藏版）》请在冰豆网上搜索。

数据挖掘实例广告投放问题eng

SASTutorial:

EnterpriseMiner

Casedescription

Supposethatyouworkforamailorderenterprisethatsendsoutacatalogoffurnishingsandhousewareseachmonth.Aspartofanupcomingsalescampaign,youwanttodistributeaspecialcatalogthatisdevotedtofinediningandcontainskitchenware,dishes,andflatware.It'stooexpensivetosendthiscatalogtoallofyourcustomers,soyouneedtotargetthosemostlikelytobuy.Youdothisbydevelopingatargetingmodelandthenusingittoproduceanewmailinglist.

Youhaveanextensiverecordofcustomerpurchases.Thedataincludesvariablesthatindicatewhethercustomersboughtkitchenware,dishes,orflatwareinthepasttwoyears.ThispurchasehistoryhasbeenusedtocreatetheCUSTDET1dataset,whichcontains49variableswiththefollowinglabels:

PurchaseDollarsSpentYearlyIncomeHomeValueOrderFrequencyRecencyMarriedNamePrefixAgeSexTelemarketInd.RentsApartmentOccupied<1YearDomesticProductApparelPurchaseLeisureProductLuxuryItems

KitchenProductDishesPurchaseFlatwarePurchaseTotalDining（kitch+dish+flat）Promo:

1-7MonthsPromo:

8-13Months$ValueperMailingCountryCodeTotalReturnsMensApparelHomeFurnitureLampsPurchaseLinensPurchaseBlanketsPurchaseTowelsPurchaseOutdoorProductCoatsPurchase

LadiesCoatsLadiesApparelHis/HerApparelJewelryPurchaseDate1stOrderTelemarketOrderAccountNumberStateCodeRaceHeatingTypeNumberofCarsNumberofKidsTravelTimeEducationLevelJobCategory

EnterpriseMiner

BasedonSASsoftware,EnterpriseMinercombinesthedataminingprocesswithgraphicaleaseofuse.Itdeliversabroadrangeofpredictiveanddescriptivemodelsthatyoucanapply,test,andcomparetodeterminethebestfitforthedata.Mostofthisisdonebymanipulatingicons:

youconnectnodesinagraphicalworkspace,adjustsettings,andruntheworkflow.Here'sanexampleoftheworkflow:

NodesareakeyconceptinEnterpriseMiner;mostofthetimeyouinteractwiththeprogrambydragginganddropping,right-clicking,ordouble-clickingthenodethatcorrespondstoaparticulartask.

1.InvokeEnterpriseMiner

AfteryoustarttheSASSystem,fromtheSASmenubar,youcanselectSolutionsAnalysisEnterpriseMiner.

Creatinganewproject

2.Tocreateanewproject,selectFileNewProject.WhentheCreatenewprojectwindowappears,enterDiningListintheNamefield.InthelocationfieldselectthebrowsebuttonandcreateanewdirectoryDiningListfortheproject.

ClicktheCreatebutton.

Youhavenowcreatedadirectoryforyourproject.EMhascreatedthreeadditionalsubdirectories:

EMDATA,EMPROJandREPORTS.

3.TheDiningListprojectappearsintheleftwindowpane.Belowitappearsthedefaultnameoftheworkflow,Untitled.SelectUntitledandenterthenewnamePropensity.

4.Gotothecoursewebsiteanddownloadthetutorialfile:

DataMining2003.sas7bdatandsaveitintheemprojsubdirectory.

Applyworkspacenodes

Youmustfirstspecifythesourceofthedataforthemailinglistproject.Todefineadatasource,youdraganddropanodeontotheworkspace.

ViewthenodesbyclickingtheToolstabatthebottomoftheleftwindowpane.Agroupedlistoficonsandlabelsappears;asmallsectionisshownbelow:

YoucanalsoselectnodesfromtheEnterpriseMinermenubaratthetopofthewindow.Thebarshowsalargerversionofsomeoftheiconsinthenodelist（Toreadthenameofanodeinthemenubar,brieflyholdthemousepointeroveritsicon.Thenamewillappearinatooltipbox）.

Defineadatasource

1.Left-clickthenodethatislabeledInputDataSourceanddragittotherightwindowpane.Releasethenodetodropitintotheworkspace.

2.Double-clickthenewnodetospecifythesourcedata.TheInputDataSourcewindowappears,withtheDatatabintheforeground.

3.ClicktheSelectbutton.TheSASDataSetwindowappears.Selecttheemprojdirectory.

4.SelectDataMining2003fromthelistoftables.ClickOKtoclosethewindow.

5.ClosetheInputDataSourcewindow.Aconfirmationboxappears.

6.ClickYestosaveyourchanges.

DataExploration

ApplytheInsightnode

EnterpriseMinerincludesanInsightnodeforexploringyourproject'sdata..Itallowsyoutoexplorethesituationofmissingvalues,outliers,orskeweddistributionscan.

1.DraganddroptheInsightnodeontotheworkspace.PlacethenewnodeundertheInputDataSourcenode

2.ConnecttheInputDataSourcenodetotheInsightnode:

1.HoldthemousepointerattheedgeoftheInputDataSourcenodeuntilitbecomesapairofcrosshairs.

2.Left-clickandquicklydragtotheInsightnode.

3.Anarrowbetweenthenodesappears.

3.Double-clicktheInsightnode.TheInsightSettingswindowappears:

4.NoticethatthedatasetnameisnotDataMining2003butanewnamethathasbeenprovidedbyEnterpriseMiner.TheDescriptionfielddisplaystheoriginaldatasetname.

5.Becauseofthelargedatastoresyoumightworkwith,theInsightnodedefaultstoexaminingasamplesizeof2,000records.InthecaseoftheDataMining2003dataset,whichhas1,966records,clicktheEntiredatasetbutton.TheSamplesizefieldwillchangetoDatasetsizetoreflectthechange.

6.ClosetheInsightSettingswindow.Aconfirmationboxappears.

7.ClickYestosaveyourchanges.

ViewInsightnodeoutput

8.Right-clicktheInsightnodeandselectRun.Agreenborderappearsaroundthenodeasitreadsyourdata,andthenaconfirmationboxpopsup.

9.ClickYestoviewresults.Atabularviewofyourdataappears:

10.Withthetableviewopen,selectAnalyzeDistributionfromtheSASSystemmenubar.

11.Awindowforselectingdistributionvariablesappears.Selectincome.ClicktheYbuttonandthenOK.Awindowwiththedistributionoftheincomevariableappears.

Createatarget–Transformingvariables

Youaretryingtotargetthebuyersofdiningwares,butthevariableDININGpresentsaproblem.BecauseitcontainsthesumofKITCHEN,DISHES,andFLATWARE,itsvaluesrangefrom0to28.Butyouarelookingforthebuyersofanydiningwares,representedbyallvaluesgreaterthan0.Whatyouneed,therefore,isabinaryversionofDINING,wherethevaluesgreaterthan0arecollapsedto1.

YoucreatevariablesinEnterpriseMinerwiththeTransformVariablesnode

1.DraganddroptheTransformVariablesnodeontotheworkspace.

2.ConnecttheInputDataSourcenodetotheTransformVariablesnode.

3.Double-clicktheTransformVariablesnode.TheTransformVariableswindowappears.

4.ClicktheCreateVariableiconontheworkspacemenubar.TheCreateVariablewindowappears.

5.EnterDINEBINintheNamefield.

6.EnterDININGNo/YesintheLabelfield.

7.ClickDefine.TheCustomizewindowappears.

8.Enterdining>0intheDINEBIN（N）=formulafieldatthebottomofthewindow.

9.ClickOK.TheCreateVariablereappearswithdining>0displayedintheFormulafield.

10.ClickOK.ThenewvariableDINEBINappearsintheTransformVariableswindow.

11.ClosetheTransformVariableswindow.Aconfirmationboxappears.

12.SelectYestosaveyourchanges

ModifyingAttributes

Younowneedtoidentifyitasthemodel'starget.ThisisdonewiththeDataSetAttributesnode

1.DraganddroptheDataSetAttributesnodeontotheworkspacetotherightoftheinputdatasourcenode.ConnecttheTransformVariablenodetotheDataSetAttributesnode.

2.Double-clicktheDataSetAttributesnode.TheDataSetAttributeswindowappears.

3.ClicktheVariablestab.

ScrolldownthelistofvariablesuntilDINEBINappears.Noticethegrayed-outcolumnModelRoleandthewhitecolumnNewModelRole.Grayed-outcolumnsreflecttheoriginaldatasetattributesandtheycannotbeedited.

Rolereferstotheuseofeachvariable.Mostvariablesaretreatedasinputvariablesinanattempttopredictthetarget.Ifyouscrolldownthelistofvariables,youwillseethatEnterpriseMinerconsiderscertainvariablesunsuitableasinputs（e.g.,dates,orvariableswithasinglevalue）.Suchvariablesaregiventherolerejected.

4.Right-clickinthecolumnNewModelRoletotherightofthevariableDINEBIN.

5.SelectSetNewModelRolefromthepop-upmenu.

6.Selecttarget.

7.Youaretryingtotargetthebuyersofdiningwares（forwhomthevariableDinebin=1）.Howeverothervariablesinthedatasetcontainthesameinformation:

KITCHEN,DISHES,andFLATWAREDINEBINhasvalueof1ifthecustomerhadboughtanydiningware.Itisthereforenecessarytoexcludethemfromtheanalysis（assigna“reject”status）.

Note:

WithintheDataSetAttributesWindow,thecolumnMeasurementreferstomeasurementlevel.Thisistherangeofvaluesthatisfoundineachvariable.Therearefivepossibleassignments:

unary-onevalue

forexample,avariablewithaparticularvaluethatwasusedtocreateadatasubset

binary-twovalues

forexample,thevariableMARITALthatcontainsNoorYes

nominal-morethantwonon-numericvalues,butnoimpliedorder

forexample,STATECODthatcontainsAK,AL,AR,AZ,etc.

ordinal-morethantwobutnotmorethantennumericvalues,withimpliedorder

forexample,NUMCARSthatcontainsvaluesfrom0to3

interval-morethantennumericvalues

forexample,AMOUNTthatcontainsmanydifferentdollarvalues

AneworderforvaluesinTargetVariable

Whenyoubuildamodel,EnterpriseMinerconsidersthetargeteventtobethefirstsortedvalueofthetargetvariable.Thedefaultsortorderisascending.Butthenewtargetvariable,DINEBIN,containsvaluesof0and1,with1representingthepurchaseofanydiningwares.ThevaluesneedtobeindescendingorderforEnterpriseMinertoaimattheintendedtarget.

Tochangetheorderofatargetvariable:

1.