Supervised and Unsupervised Learning

《Supervised and Unsupervised Learning》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、SupervisedandUnsupervisedLearningKwok-LeungTsuiSystemsEngineering&EngineeringManagementCityUniversityofHongKong1/31/20121DataMining(KDD)ProcessDetermineBusinessObjectivesDataPreparationMining&ModelingConsolidationandApplication1/31/20122DataMiningandM

2、odeling1/31/20123DataMining&ModelingStartConsiderChooseModelsAlternateModelsTrainDataBuild/FitModelCollectSamplemoreDataValidationRefine/TuneModeldataData(modelsize&diagnosis)EvaluateModelTestData(e.g.Predictionerror)(EvaluationData)NOMeetaccuracyreqt

3、.YESScoreDataPredictionMakeDecisions1/31/20124DataDescription&Visualization•Descriptivestatisticalmeasures–Centraltendency/location–Dispersion/spread–Shape&symmetry•ClassCharacterizationandComparisons–Analyticalcharacterization–Attributerelevanceanaly

4、sis–Classdiscriminationandcomparisons•DataVisualization–Scatter-plotmatrix&densityplot–3-Dstereoscopicscatter-plot–Parallelcoordinateplot1/31/20125Supervised&UnsupervisedLearning•Supervisedlearning:–Learningwithateacher–Classification,e.g.onlineshoppe

5、rs(buyersVs.non-buyers)•Unsupervisedlearning:–Learningwithoutateacher–Clustering,e.g.onlineshoppers(segmentationofnon-buyers)•Otherrelatedterms:–MachineLearning(analogiestohumanreceiving)–NeuralNetworks(biologicalanalogiestobrain)1/31/20126SupervisedL

6、earning•Inputs:(Predictors,independentvariables,y)–Asetofvariableswhicharemeasuredorpreset.•Outputs:(Responses,dependentvariables,x)–Asetofmeasurablevariableswhichareinfluencedbytheinputs•Steps:–Establishmodels/systems(yhat)basedoncollectedinputs&outp

7、uts(xandy).–Predictthevaluesofoutputsbasedontheestablishedmodels/systemsandanewsetofspecifiedinputs.1/31/20127SupervisedLearning•Learningwithateacher(generalization)–Studentpresentsanswer(givenxyii)–Teacherprovidesthecorrectansweryioranerrorforstuden

8、t’sanswer–Theresultischaracterizedbysomelossfunction:L(y,y)–Objective:Minimizetheexpectedloss•Functionapproximation:Y=f(x,)1/31/20128ProblemsinSupervisedLearning•(Application/ProblemOriented)–Classificationproblem:Outputiscategorical/qualita