模式识别--聚类分析.ppt

《模式识别--聚类分析.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、聚类分析聚类的概念基于k质心的聚类方法K-means方法K-means的性质C-means层次聚类2011/5/121樊明锁聚类分析ClusterAnalysis(聚类分析）Findinggroupsofobjectssuchthattheobjectsinagroupwillbesimilar(orrelated)tooneanotheranddifferentfrom(orunrelatedto)theobjectsinothergroupsInter-clusterdistancesaremaximizedIntra-clusterdistancesar

2、eminimized2011/5/122樊明锁聚类分析ApplicationsofClusterAnalysisUnderstandingGrouprelateddocumentsforbrowsing,groupgenesandproteinsthathavesimilarfunctionality,orgroupstockswithsimilarpricefluctuationsSummarizationReducethesizeoflargedatasetsClusteringprecipitationinAustralia2011/5/123樊明锁聚类

3、分析WhatisNOTClusterAnalysis?SupervisedclassificationHaveclasslabelinformationSimplesegmentationDividingstudentsintodifferentregistrationgroupsalphabetically,bylastnameResultsofaqueryGroupingsarearesultofanexternalspecificationGraphpartitioningSomemutualrelevanceandsynergy,butareasare

4、notidentical2011/5/124樊明锁聚类分析NotionofaClustercanbeAmbiguousHowmanyclusters?FourClustersTwoClustersSixClusters2011/5/125樊明锁聚类分析TypesofClusteringsAclusteringisasetofclustersImportantdistinctionbetweenhierarchicalandpartitionalsetsofclustersPartitionalClustering(flat)Adivisiondataobjec

5、tsintonon-overlappingsubsets(clusters)suchthateachdataobjectisinexactlyonesubsetHierarchicalclusteringAsetofnestedclustersorganizedasahierarchicaltree2011/5/126樊明锁聚类分析PartitionalClusteringOriginalPointsAPartitionalClustering2011/5/127樊明锁聚类分析HierarchicalClusteringTraditionalHierarchi

6、calClusteringNon-traditionalHierarchicalClusteringNon-traditionalDendrogramTraditionalDendrogram2011/5/128樊明锁聚类分析TypesofClusters:Well-SeparatedWell-SeparatedClusters:Aclusterisasetofpointssuchthatanypointinaclusteriscloser(ormoresimilar)toeveryotherpointintheclusterthantoanypointnot

7、inthecluster.3well-separatedclusters2011/5/129樊明锁聚类分析TypesofClusters:Center-BasedCenter-basedAclusterisasetofobjectssuchthatanobjectinaclusteriscloser(moresimilar)tothe“center”ofacluster,thantothecenterofanyotherclusterThecenterofaclusterisoftenacentroid,theaverageofallthepointsinth

8、ecluster,oramedoid,