%0 Journal Article
%T 分类分析中基于信息论准则的特征选取
%A 黄金杰
%A 吕宁
%A 李双全
%A 蔡云泽
%J 自动化学报
%P 383-392
%D 2008
%R 10.3724/SP.J.1004.2008.00383
%X ？Featureselectionaimstoreducethedimensionalityofpatternsforclassificatoryanalysisbyselectingthemostinformativeinsteadofirrelevantand/orredundantfeatures.Inthisstudy,twonovelinformation-theoreticmeasuresforfeaturerankingarepresented:oneisanimprovedformulatoestimatetheconditionalmutualinformationbetweenthecandidatefeaturefiandthetargetclassCgiventhesubsetofselectedfeaturesS,i.e.,I(C;fi|S),undertheassumptionthatinformationoffeaturesisdistributeduniformly;theotherisamutualinformation(MI)basedconstructivecriterionthatisabletocapturebothirrelevantandredundantinputfeaturesunderarbitrarydistributionsofinformationoffeatures.Withthesetwomeasures,twonewfeatureselectionalgorithms,calledthequadraticMI-basedfeatureselection(QMIFS)approachandtheMI-basedconstructivecriterion(MICC)approach,respectively,areproposed,inwhichnoparameterslikeβinBattiti'sMIFSand(KwakandChoi)'sMIFS-Umethodsneedtobepreset.Thus,theintractableproblemofhowtochooseanappropriatevalueforβtodothetradeoffbetweentherelevancetothetargetclassesandtheredundancywiththealready-selectedfeaturesisavoidedcompletely.ExperimentalresultsdemonstratethegoodperformancesofQMIFSandMICConbothsyntheticandbenchmarkdatasets.
%K Patternclassification
%K datamining
%K featureselection
%K information-theoreticmeasures
%U http://www.aas.net.cn/CN/abstract/abstract15906.shtml