針對非均勻數(shù)據(jù)集的DBSCAN過濾式改進算法.pdf

1、重慶大學碩士學位論文針對非均勻數(shù)據(jù)集的DBSCAN過濾式改進算法姓名：吳林敏申請學位級別：碩士專業(yè)：計算機系統(tǒng)結(jié)構(gòu)指導教師：熊忠陽20090601重慶大學碩十學位論文英文摘要ABSTRACTDataMiningfindsoutconnotatiVe，unknownandpotentiallyValua_bleknowledgeaIldmlesClusteringisoneoftheimportantresearchfieldsindat

2、aminingClusteringistheprocessof伊oupingphysicalorabstractsetsintoseVeralsimilarclustersTheclustersproducedbyclusteringaresetsofdataobjectsOneobjectissimilartotheotherobjectsinthesarnecluster，andisdifI’erentfromtheobjectsi

3、ndifferentclustersIIlmanyapplications，theobjectsinthesanleclustercaIlbetreatedasawholeWhenanalyzingabig，complicated，continuousdatabaseOrtotallyunknownstructures，clusteringisaVeryusefult001Atpresent，clusteringanalysisalgo

4、rithmcanbesortedintoseVerall(inds：p礬itionmethod，llierarchymethod，densitybasedmethod，griddingbasedmethodandmodelbasedmethodDBSCANalgorithmisatypicaldensitybasedmethodThemeritsofDBSCAN砌ethatitcanfindsout抽itraryshapecluster

5、s，anditsclusteringresultish鋤dlyinnuencedbynoisepointsTheshonpointsofDBSCAN孤elistedasf01lows：First，iftheValuesoftheseglobalVariablesarenotappropriate，thenclusteringresultcouldbeinfluencedSecond，whenthedistributionofdatase

6、tsisuneVen，theclusteringqualityisVerypoorForthedisadVantagesofDBSCAN，theDBSCANalgorithmbasedonfiltrationaJgorit№isproposedAtthesa玎1etime，thealgorithmalsocoulddecreasetheregionqueryFirst，thealgorithmcalculatesthek—distoft

7、hedatasets，andthenuses1dimensionclusteringtogetalltheclustersARerthenoiseclustersareremoVed，theclustersthatcanrepresenttheprimaurydensitiescouldbeobtainedSecond，theimproVedalgoritIlIIlgetsseVeralValuesofparameterZ≯sfordi

8、f佗rentdensities，andproVidesforthefiltrationAftergotpar鋤eterZ≯研，theimproVedalgorithmusesdif億rentValuesof上≯研toclusterthedatasets，andthenfindoutclustersfbrthedatasetsForthenextprocess，thepointsthathaVebeenclusteredareignore

9、d，whichavoidsm砌dngbothdenserareasandsparseronesasoneclusterTheimproVedalgoritllIIlusesl—dimensionclusteringtoget2≯毋，andthenusesdif億rentValuesof正≯研toclusterSo，whenthedistributionofdatasetsisuneven，theclusteringresultsofim