资源描述：
《An Introduction to Variable and Feature Selection》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、JournalofMachineLearningResearch3(2003)1157-1182Submitted11/02;Published3/03AnIntroductiontoVariableandFeatureSelectionIsabelleGuyonISABELLE@CLOPINET.COMClopinet955CrestonRoadBerkeley,CA94708-1501,USAAndreElisseeff´ANDRE@TUEBINGEN.MPG.DEEmpiricalInfe
2、renceforMachineLearningandPerceptionDepartmentMaxPlanckInstituteforBiologicalCyberneticsSpemannstrasse3872076Tubingen,Germany¨Editor:LesliePackKaelblingAbstractVariableandfeatureselectionhavebecomethefocusofmuchresearchinareasofapplicationforwhichdat
3、asetswithtensorhundredsofthousandsofvariablesareavailable.Theseareasincludetextprocessingofinternetdocuments,geneexpressionarrayanalysis,andcombinatorialchemistry.Theobjectiveofvariableselectionisthree-fold:improvingthepredictionperformanceofthepre-d
4、ictors,providingfasterandmorecost-effectivepredictors,andprovidingabetterunderstandingoftheunderlyingprocessthatgeneratedthedata.Thecontributionsofthisspecialissuecoverawiderangeofaspectsofsuchproblems:providingabetterdefinitionoftheobjectivefunction,
5、featureconstruction,featureranking,multivariatefeatureselection,efficientsearchmethods,andfeaturevalidityassessmentmethods.Keywords:Variableselection,featureselection,spacedimensionalityreduction,patterndiscov-ery,filters,wrappers,clustering,informatio
6、ntheory,supportvectormachines,modelselection,statisticaltesting,bioinformatics,computationalbiology,geneexpression,microarray,genomics,proteomics,QSAR,textclassification,informationretrieval.1IntroductionAsof1997,whenaspecialissueonrelevanceincludings
7、everalpapersonvariableandfeatureselectionwaspublished(BlumandLangley,1997,KohaviandJohn,1997),fewdomainsexploredusedmorethan40features.Thesituationhaschangedconsiderablyinthepastfewyearsand,inthisspecialissue,mostpapersexploredomainswithhundredstoten
8、softhousandsofvariablesorfeatures:1Newtechniquesareproposedtoaddressthesechallengingtasksinvolvingmanyirrelevantandredundantvariablesandoftencomparablyfewtrainingexamples.Twoexamplesaretypicalofthenewapplicationdomainsandserveusasillustrationthrougho
