资源描述：
《Feature Selection for Nonlinear Regression and its Application to SDM_0533-000057》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、FeatureSelectionforNonlinearRegressionanditsApplicationtoCancerResearchYijunSunJinYaoySteveGoodisonzAbstracttheissueistoperformfeatureselectiontoextracttheFeatureselectionisafundamentalprobleminmachinemostrelevantinformationabouteachobserveddatumlearning.Withtheadventofh
2、igh-throughputtechnolo-fromapotentiallyoverwhelmingquantityofitsfeaturesgies,itbecomesincreasinglyimportantinawiderange[7].Anexamplewherefeatureselectionplaysacriticalofscienti cdisciplines.Inthispaper,weconsidertheroleistheuseofoligonucleotidemicroarrayfortheiden-problem
3、offeatureselectionforhigh-dimensionalnon-ti cationofcancer-associatedgeneexpressionpro lesoflinearregression.Thisproblemhasnotyetbeenwellad-prognosticvalue.Typically,thenumberofsamplesisdressedinthecommunity,andexistingmethodssu eraroundonehundred,whilethenumberofgenesass
4、oci-fromissuessuchaslocalminima,simpli edmodelas-atedwithrawdataisontheorderofthousandsorevensumptions,highcomputationalcomplexityandselectedtensofthousands.Theidenti cationofasmallfrac-featuresnotdirectlyrelatedtolearningaccuracy.Wetionofgenesthatdrivecanceroustumorgrowt
5、hand/orproposeanewwrappermethodthataddressessomeofspreadcansigni cantlyimprovetheaccuracyofcancertheseissues.Westartbydevelopinganewapproachprognosis.Inadditiontodefyingthecurseofdimen-toestimatingsampleresponsesandpredictionerrors,sionality,eliminatingirrelevantfeaturesc
6、analsoreduceandthendeployafeatureweightingstrategyto ndaprocessingtimeofdataanalysisandthecostofcollect-featuresubspacewhereapredictionerrorfunctionisingirrelevantfeatures.Inmanycases,featureselectionminimized.Weformulateitasanoptimizationprob-canalsoprovidesigni cantinsi
7、ghtsintothenatureoflemwithintheSVMframeworkandsolveitusingantheproblemunderinvestigation.iterativeapproach.Ineachiteration,agradientdescentTheproblemoffeatureselectionhasbeenex-basedapproachisderivedtoeciently ndasolution.tensivelystudiedinthemachinelearningcommunityAlar
8、ge-scalesimulationstudyisperformedonfoursyn-[11,7,23,24,25].However,themajorityoftheworkthetican
