第十九章-聚类分析-(Clustering-Analysis)---中南大学.ppt

《第十九章-聚类分析-(Clustering-Analysis)---中南大学.ppt》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、Chapter19ClusteringAnalysisContentSimilaritycoefficientHierarchicalclusteringanalysisDynamicclusteringanalysisOrderedsampleclusteringanalysisDiscriminantAnalysis:havingknownwithcertaintytocomefromtwoormorepopulations,it’samethodtoacquirethediscriminatemodelthatwillallocatefurtherin

2、dividualstothecorrectpopulation.ClusteringAnalysis:astatisticmethodforgroupingobjectsofrandomkindintorespectivecategories.It’susedwhenthere’snopriorihypotheses,buttryingtofindthemostappropriatesortingmethodresortingtomathematicalstatisticsandsomecollectedinformation.Ithasbecomethef

3、irstselectedmeanstouncovergreatcapacityofgeneticmessages.Botharemethodsofmultivariatestatisticstostudyclassification.Clusteringanalysisisamethodofexploringstatisticalanalysis.Itcanbeclassifiedintotwomajorspeciesaccordingtoitsaims.Forexample,mreferstothenumberofvariables(i.e.indexe

4、s)whilenreferstothatofcases(i.e.samples),youcandoasfollows:(1)R-typeclustering:alsocalledindexclustering.Themethodtosortthemkindsofindexes,aimingatloweringthedimensionofindexesandchoosingtypicalones.（2）Q-typeclustering:alsocalledsampleclustering.Themethodtosortthenkindsofsamplestof

5、indthecommonnessamongthem.ThemostimportantthingforbothR-typeclusteringandQ-typeclusteringisthedefinitionofsimilarity,thatishowtoquantifysimilarity.Thefirststepofclusteringistodefinethemetricsimilaritybetweentwoindexesortwosamples-similaritycoefficient§1similaritycoefficient1similar

6、itycoefficientofR-typeclusteringSupposetherearemkindsofvariables:X1，X2，…，Xm.R-typeclusteringusuallyusetheabsolutevalueofsimplecorrelationcoefficienttodefinethesimilaritycoefficientamongvariables:Thetwovariablestendtobemoresimilarwhentheabsolutevalueincreases.Similarly,Spearmanrankc

7、orrelationcoefficientcanbeusedtodefinethesimilaritycoefficientofnon-normalvariables.Butwhenthevariablesareallqualitativevariables,it’sbesttousecontingencycoefficient.2．SimilaritycoefficientcommonlyusedinQ-typeclustering:Supposetherearencasesregardasnspotsinamdimensionsspace,distanc

8、ebetweentwospotscanbeusedt