Cluster analysis using R 怎样用R来做类别分析

《Cluster analysis using R 怎样用R来做类别分析》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、CLUSTERANALYSISStevenM.Ho!andDepartmentofGeology,UniversityofGeorgia,Athens,GA30602-2501January2006IntroductionClusteranalysisincludesabroadsuiteoftechniquesdesignedtofindgroupsofsimilaritemswithinadataset.Partitioningmethodsdividethedatasetintoanumberofgroupspre-designatedbytheus

2、er.Hierarchicalclustermethodsproduceahierarchyofclustersfromsmallclustersofverysimilaritemstolargeclustersthatincludemoredissimilaritems.Hier-archicalmethodsusuallyproduceagraphicaloutputknownasadendrogramortreethatshowsthishierarchicalclusteringstructure.Somehierarchicalmethodsa

3、redivisive,thatprogressivelydividetheonelargeclustercomprisingallofthedataintotwosmallerclustersandrepeatthisprocessuntilallclustershavebeendivided.Otherhierarchicalmethodsareagglomerativeandworkintheoppositedirectionbyfirstfindingtheclustersofthemostsimilaritemsandprogressivelyadd

4、inglesssimilaritemsuntilallitemshavebeenincludedintoasinglelargecluster.ClusteranalysiscanberunintheQ-modeinwhichclustersofsamplesaresoughtorintheR-mode,whereclustersofvariablesaredesired.Hierarchicalmethodsareparticularlyusefulinthattheyarenotlimitedtoapre-determinednumberofclus

5、tersandcandisplaysimilarityofsamplesacrossawiderangeofscales.Ag-glomerativehierarchicalmethodsareparticularlycommoninthenaturalsciencesandtheywillbethefocusofthislecture.ComputationAswithothermultivariatemethods,thestartingpointisadatamatrixconsistingofnrowsofsamplesandpcolumnsof

6、variables,calledannxp(nbyp)matrix.Hierarchicalagglomera-tiveclusteranalysisbeginsbycalculatingamatrixofdistancesamongitemsinthisdatama-trix.AlthoughclusteranalysiscanberunintheR-modewhenseekingrelationshipsamongvariables,thisdiscussionwillassumethataQ-modeanalysisisbeingrun.InQ-m

7、odeanaly-sis,thedistancematrixisasquare,symmetricmatrixofsizenxnthatexpressesallpossiblepairwisedistancesamongsamples.Manydistancemetricscanbeused.Beforeclusteringhasbegun,eachsampleisconsideredagroup,albeitofasinglesample.Clusteringbeginsbyfindingthetwogroupsthataremostsimilar,ba

8、sedonthedistancema-trix,andmergingthemin