数据挖掘导论第4课数据分类和预测.ppt

《数据挖掘导论第4课数据分类和预测.ppt》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、第4课数据分类和预测徐从富，副教授浙江大学人工智能研究所浙江大学本科生《数据挖掘导论》课件内容提纲Whatisclassification?Whatisprediction?IssuesregardingclassificationandpredictionClassificationbydecisiontreeinductionBayesianClassificationPredictionSummaryReferenceClassificationpredictscategoricalclasslabels(discr

2、eteornominal)classifiesdata(constructsamodel)basedonthetrainingsetandthevalues(classlabels)inaclassifyingattributeandusesitinclassifyingnewdataPredictionmodelscontinuous-valuedfunctions,i.e.,predictsunknownormissingvaluesTypicalapplicationsCreditapprovalTargetmark

3、etingMedicaldiagnosisFrauddetectionClassificationvs.PredictionClassification—ATwo-StepProcessModelconstruction:describingasetofpredeterminedclassesEachtuple/sampleisassumedtobelongtoapredefinedclass,asdeterminedbytheclasslabelattributeThesetoftuplesusedformodelcon

4、structionistrainingsetThemodelisrepresentedasclassificationrules,decisiontrees,ormathematicalformulaeModelusage:forclassifyingfutureorunknownobjectsEstimateaccuracyofthemodelTheknownlabeloftestsampleiscomparedwiththeclassifiedresultfromthemodelAccuracyrateistheper

5、centageoftestsetsamplesthatarecorrectlyclassifiedbythemodelTestsetisindependentoftrainingset,otherwiseover-fittingwilloccurIftheaccuracyisacceptable,usethemodeltoclassifydatatupleswhoseclasslabelsarenotknownClassificationProcess(1):ModelConstructionTrainingDataCla

6、ssificationAlgorithmsIFrank=‘professor’ORyears>6THENtenured=‘yes’Classifier(Model)ClassificationProcess(2):UsetheModelinPredictionClassifierTestingDataUnseenData(Jeff,Professor,4)Tenured?Supervisedvs.UnsupervisedLearningSupervisedlearning(classification)Supervisio

7、n:Thetrainingdata(observations,measurements,etc.)areaccompaniedbylabelsindicatingtheclassoftheobservationsNewdataisclassifiedbasedonthetrainingsetUnsupervisedlearning(clustering)TheclasslabelsoftrainingdataisunknownGivenasetofmeasurements,observations,etc.withthea

8、imofestablishingtheexistenceofclassesorclustersinthedataIssuesRegardingClassificationandPrediction(1):DataPreparationDatacleaningPreprocessdatainorderto