线性回归模型在SAS_EM中的应用实例.doc

《线性回归模型在SAS_EM中的应用实例.doc》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、......Chapter3PredictiveModelingUsingRegression3.1IntroductiontoRegression3-33.2RegressioninEnterpriseMiner3-8.专业资料.......Chapter3PredictiveModelingUsingRegression3.1IntroductiontoRegression3-33.2RegressioninEnterpriseMiner3-8.专业资料.......3.1IntroductiontoRegression.专业资料.......TheRegres

2、sionnodeinEnterpriseMinerdoeseitherlinearorlogisticregressiondependinguponthemeasurementlevelofthetargetvariable.Linearregressionisdoneifthetargetvariableisanintervalvariable.Inlinearregressionthemodelpredictsthemeanofthetargetvariableatthegivenvaluesoftheinputvariables.Logisticregress

3、ionisdoneifthetargetvariableisadiscretevariable.Inlogisticregressionthemodelpredictstheprobabilityofaparticularlevel(s)ofthetargetvariableatthegivenvaluesoftheinputvariables.Becausethepredictionsareprobabilities,whichareboundedby0and1andarenotlinearinthisspace,theprobabilitiesmustbetra

4、nsformedinordertobeadequatelymodeled.Themostcommontransformationforabinarytargetisthelogittransformation.Probitandcomplementarylog-logtransformationsarealsoavailableintheregressionnode..专业资料.......Recallthatoneassumptionoflogisticregressionisthatthelogittransformationoftheprobabilities

5、ofthetargetvariableresultsinalinearrelationshipwiththeinputvariables..专业资料.......Regressionusesonlyfullcasesinthemodel.Thismeansthatanycase,orobservation,thathasamissingvaluewillbeexcludedfromconsiderationwhenbuildingthemodel.Asdiscussedearlier,whentherearemanypotentialinputvariablesto

6、beconsidered,thiscouldresultinanunacceptablyhighlossofdata.Therefore,whenpossible,missingvaluesshouldbeimputedpriortorunningaregressionmodel.Otherreasonsforimputingmissingvaluesincludethefollowing:·Decisiontreeshandlemissingvaluesdirectly,whereasregressionandneuralnetworkmodelsignoreal

7、lobservationswithmissingvaluesonanyoftheinputvariables.Itismoreappropriatetocomparemodelsbuiltonthesamesetofobservations.Therefore,beforedoingaregressionorbuildinganeuralnetworkmodel,youshouldperformdatareplacement,particularlyifyouplantocomparetheresultstoresultsobtainedfromadecisio