Cluster Analysis R的用法

《Cluster Analysis R的用法》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、ClusterAnalysisTheTwoStepClusterAnalysisprocedureisanexploratorytooldesignedtorevealnaturalgroupings(orclusters)withinadatasetthatwouldotherwisenotbeapparent.Thealgorithmemployedbythisprocedurehasseveraldesirablefeaturesthatdifferentiateitfromtraditionalclusteringtechniques:·Theabilitytocreat

2、eclustersbasedonbothcategoricalandcontinuousvariables.·Automaticselectionofthenumberofclusters.·Theabilitytoanalyzelargedatafilesefficiently.Inordertohandlecategoricalandcontinuousvariables,theTwoStepClusterAnalysisprocedureusesalikelihooddistancemeasurewhichassumesthatvariablesintheclustermo

3、delareindependent.Further,eachcontinuousvariableisassumedtohaveanormal(Gaussian)distributionandeachcategoricalvariableisassumedtohaveamultinomialdistribution.Empiricalinternaltestingindicatesthattheprocedureisfairlyrobusttoviolationsofboththeassumptionofindependenceandthedistributionalassumpt

4、ions,butyoushouldtrytobeawareofhowwelltheseassumptionsaremet.ThetwostepsoftheTwoStepClusterAnalysisprocedure'salgorithmcanbesummarizedasfollows:Step1.TheprocedurebeginswiththeconstructionofaClusterFeatures(CF)Tree.Thetreebeginsbyplacingthefirstcaseattherootofthetreeinaleafnodethatcontainsvari

5、ableinformationaboutthatcase.Eachsuccessivecaseisthenaddedtoanexistingnodeorformsanewnode,baseduponitssimilaritytoexistingnodesandusingthedistancemeasureasthesimilaritycriterion.Anodethatcontainsmultiplecasescontainsasummaryofvariableinformationaboutthosecases.Thus,theCFtreeprovidesacapsulesu

6、mmaryofthedatafile.Step2.TheleafnodesoftheCFtreearethengroupedusinganagglomerativeclusteringalgorithm.Theagglomerativeclusteringcanbeusedtoproducearangeofsolutions.Todeterminewhichnumberofclustersis"best",eachoftheseclustersolutionsiscomparedusingSchwarz'sBayesianCriterion(BIC)ortheAkaikeInfo

7、rmationCriterion(AIC)astheclusteringcriterion.Carmanufacturersneedtobeabletoappraisethecurrentmarkettodeterminethelikelycompetitionfortheirvehicles.Ifcarscanbegroupedaccordingtoavailabledata,thistaskcanbelargelyautomaticusingclusteranalysis.S