classification1-full

《classification1-full》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Classification(Part1)1OverviewBasicdecisiontreeclassifier(DT)constructionSometechnicalissuesofDTclassificationEvaluatingclassifiers2ClassificationWhatcommoncharacteristicsaresharedbytheredpeople,andnotbythepurplepeople?test(model)badcredithistorygoodcredithistory3RecordAgeInc

2、omeStudenCredit-Own-idtratingcomputer1<30HighNoBadNo2<30HighNoGoodNo330..HighNoBadYes404>40MediumNoBadYes5>40LowYesBadYes6>40LowYesGoodNo730..LowYesGoodYes408<30MediumNoBadNo9<30LowYesBadYes10>40MediumYesBadYes11<30MediumYesGoodYes1230..MediumNoGoodYes401330..HighYesBadYes4014>4

3、0MediumNoGoodNo4ExamplerulesIfage<30andisnotastudentnotacomputerownerIfage<30andisastudentacomputerownerIfageisbetween30to40acomputerownerIfage>40withagoodcreditratingnotacomputerownerIfage>40withabadcreditratingacomputerowner5DataModeladatasetconsistsofanumberofrecor

4、dseachrecordconsistsofanumberofattributevaluesoneparticularattributeiscalledthelabel(orclass)recordsthatsharethesamelabelvalueformaclasswewanttodiscoverrulesthathelppredict,givenafuture(unclassified)record,whichclasstherecordshouldbelong6SupervisedLearningourgeneralapproach

5、totheclassificationproblemprepareadatasetoflabeledrecordsdrawarandomsample(e.g.,80%),callitthetrainingsetusethetrainingsettotrainaclassifierapplytheclassifiertotherest(e.g.,20%)oftherecords(thetestset)toevaluatetheaccuracyoftheclassifierduringthetrainingphase,theclassifieri

6、sfedwithlabeledrecords(examplesofeachclass),thelearningissupervised.Thismachinelearningapproachiscalledsupervisedlearning.7Decision-TreeClassifier(DT)weconsiderhowtobuildadecision-treeclassifierweassumethatattributesareallnominal–theyonlytakeonafinitesetofnamed-valuesnumerica

7、ttributescanbemappedtonominalattributesbydiscretizationorbinningexample:incomecanbegroupedintolow(<80K),medium(>=80K,<250K),high(>=250K)aperson’sheightcanbeveryshort,short,average,tall,verytall(withasuitablemapping)8Decision-TreeClassifieradecisiontreeisatreestructureprop

8、erties:eachinternalnodedenotesatestonanattribu