摘要
Theperformanceoftheclassicalclusteringalgorithmisnotalwayssatisfiedwiththehigh-dimensionaldatasets,whichmakeclusteringmethodlimitedinmanyapplication.Tosolvethisproblem,clusteringmethodwithProjectionPursuitdimensionreductionbasedonImmuneClonalSelectionAlgorithm(ICSA-PP)isproposedinthispaper.ProjectionpursuitstrategycanmaintainconsistentEuclideandistancesbetweenpointsinthelow-dimensionalembeddingswheretheICSAisusedtosearchoptimizingprojectiondirection.Theproposedalgorithmcanconvergequicklywithlessiterationtoreducedimensionofsomehigh-dimensionaldatasets,andinwhichspace,K-meanclusteringalgorithmisusedtopartitionthereduceddata.TheexperimentresultsonUCIdatashowthatthepresentedmethodcansearchquickertooptimizeprojectiondirectionthanGeneticAlgorithm(GA)andithasbetterclusteringresultscomparedwithtraditionallineardimensionreductionmethodforPrincipleComponentAnalysis(PCA).
出版日期
2010年04月14日(中国Betway体育网页登陆平台首次上网日期,不代表论文的发表时间)