欢迎来到天天文库
浏览记录
ID:39145632
大小:1.93 MB
页数:51页
时间:2019-06-25
《组特征选择算法与其应用分析》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、Abst髓lctAbStraCtClaSsificationmodelsandfIeatureselectionmetllodsongeneexpressionmicroarraydatainbioinfomaticsarenowthefocusofattentioniIlmachineleaminganddatamillingcommulli吼The仇nhmatgenesarelikelytoco—regulatein缸lctioningmeaJlsthatf.eature铲oupsdoexistiIlmic
2、roarraydata,anddomaillexpertswouldliketoexploittllem.Therefore,t11isp印erstudiesthegroupfeatureselectionalgorithms,t巧ingtofindthef.eanlregroupsWhjchproVidemoreinsightsintotlleuIlderlyingdatarelationshipsfordomainexperts,andmeanWhjlereducethedimensionalit),and
3、improVeclaLssificationaccu姐cy.Generally,伊oupfeatureselectionalgoritllmscanbediVidedintotwocategoriesaccordingtothewayof且ndingfIeaturegroups:theexplicitonesaIldtlleimplicitones.111anexplicitgroupf-e狐Ⅱ℃selectionalgoritlu:11,groupsarefbuIlduIldersomecriterion,a
4、Ildf.eaturesinthes锄e掣oupareaLlwaysKghlycorrelated.ARermegroupsareidentified,t11ef.e咖reselectionalgorimmispe怕medonr印resentatiVefea:tures丘omeachgrouptoobtainthefinalgroupoffeanlres.BycontraSt,theimplicit伊oupfea_tureselectionalgorithmsdonotfind铲oupsdirectly,wKl
5、e伊oupscanbeidentifiedbytheresultsmatmef.eatureselectionalgorithmproVides.Forthese撕osortsof掣oupf.ea.tlJreselectionalgorimms,t11emaillcontributionsoftllispaperare:1.Proposeafbat吡eclusteringbasedexplicitgroupf.eatureselectionalgorithm,FC-gRFE(FeatureClustemgbaS
6、edGroupSVM-RFE).111isalgori廿Hnpe疵msf.e栅eclusteringontlletI-aillingsettofindt11efeature铲oups,andmenselectf.eaturesatthegroupleVelViaSVMRFE.Inaddition,SW-gRFE(S锄pleWeightingFC-gRFE)砌chbaSedonFC-gRFEwi也s锄pleweightingisproposed.T11is“goritllIIlfirstlycalculatess
7、锄pleweightsaccordingtos锄pleimportanceandmenperfomsFC-gRFEtogetgroupfeatureselectionresultsonmeweighteds锄pleset.ExperimentalresultsonmiCroarraydatashowthattheproposedalgoritllmcanfindthef.eanlre伊oupsofdata,谢thoutsc撕母ingtheclaSsificationaccuracy.2.Proposeanimp
8、licitgroupfeatureselectionalgoritlunwithmodifiedweights,CW二groupS(Coe伍cientWeightgroupfIeamreSelection).111ealgori‰firstlycodeseachf.eatureonmeoriginalfea_turesetViaaSparsemodel,E1aSticNet,andt1
此文档下载收益归作者所有