数据挖掘之Cluster Analysis课件.ppt

《数据挖掘之Cluster Analysis课件.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、ClusterAnalysisoutlineWhatIsClusterAnalysis?TypesofDatainClusterAnalysisACategorizationofMajorClusteringMethodsTypicalClusteringMethodsOutlierAnalysisWhatisClusterAnalysis?ClusteringistheprocessofgroupingthedataintoclassesorclusterssothatobjectsSimil

2、artooneanotherwithinthesameclusterDissimilartotheobjectsinotherclustersExamplesofClusteringApplicationsTextclustering：websearchenginesMarketing:HelpmarketersdiscoverdistinctgroupsintheircustomerbasesInsurance:Identifyinggroupsofmotorinsurancepolicyho

3、lderswithahighaverageclaimcostLanduse:IdentificationofareasofsimilarlanduseinanearthobservationdatabaseCity-planning:Identifyinggroupsofhousesaccordingtotheirhousetype,value,andgeographicallocationoutlineWhatIsClusterAnalysis?TypesofDatainClusterAnal

4、ysisACategorizationofMajorClusteringMethodsTypicalClusteringMethodsOutlierAnalysisDataStructuresDatamatrixDataStructuresDissimilaritymatrixd(i,j)isthemeasureddifferenceordissimilaritybetweenobjectiandobjectj,anonnegativenumber.Typeofdatainclusteringa

5、nalysisInterval-scaledvariablesBinarydiscretevariablesMulti-valueddiscreteattributesInterval-scaledvariablesinterval-scaledvariablesarecontinuousmeasurements.suchasdistance、age、income、weightStandardization:differentunitdifferentclusteringstructureSim

6、ilarityandDissimilarityBetweenObjectsDistancesarenormallyusedtomeasurethesimilarityordissimilaritybetweentwodataobjectsSomepopularonesinclude:MinkowskidistanceManhattandistanceEuclideandistanceMinkowskidistance:wherei=(xi1,xi2,…,xip)andj=(xj1,xj2,…,x

7、jp)aretwop-dimensionaldataobjects,andqisapositiveintegerIfq=1,disManhattandistance:Ifq=2,disEuclideandistance:BinarydiscretevariablesAbinaryvariablehasonlytwostates:0or1.Symmetric（对称的）:genderAsymmetric:diseasetestDissimilaritybetweenBinaryvariablesOb

8、jectiObjectjaisthenumberofvariablesthatequal1forbothobjectiandj.Pisthetotalnumberofvariables,p=a+b+c+d.DissimilaritybetweenBinaryVariablesExamplegenderisasymmetricattributetheremainingattributesareasymmetricbinaryletthevaluesYandPbesetto1,andthevalue