文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > 计算机软件及应用 > Improved Feature Selection Algorithm Based on SVM and Correlation
侵权投诉

Improved Feature Selection Algorithm Based on SVM and Correlation

ImprovedFeatureSelectionAlgorithmBasedon

SVMandCorrelation

Zong-XiaXie,Qing-HuaHu,andDa-RenYu

HarbinInstituteofTechnology,Harbin,P.R.China

xiezongxia@hcms.hit.edu.cn

Abstract.Asafeatureselectionmethod,supportvectormachines-recursivefeatureelimination(SVM-RFE)canremoveirrelevancefea-turesbutdon’ttakeredundantfeaturesintoconsideration.Inthispa-per,itisshownwhythismethodcan’tremoveredundantfeaturesandanimprovedtechniqueispresented.Correlationcoe cientisintroducedtomeasuretheredundancyintheselectedsubsetwithSVM-RFE.Thefeatureswhichhaveagreatcorrelationcoe cientwithsomeimportantfeatureareremoved.ExperimentalresultsshowthatthereactuallyareseveralstronglyredundantfeaturesintheselectedsubsetsbySVM-RFE.Thecoe cientsarehighto0.99.Theproposedmethodcannotonlyre-ducethenumberoffeatures,butalsokeeptheclassi cationaccuracy.1Introduction

Inrecentyears,datahasbecomeincreasinglylargernotonlyinnumberofin-stancesbutalsoinnumberoffeaturesinmanyapplications,suchasgeneselec-tionfrommicro-arraydataandtextautomaticcategorization,etc.,wherethenumberoffeaturesintherawdatarangesfromhundredstotensofthousands.Highdimensionalitybringsgreatdi cultyinunderstandingthedata,construct-ingclassi ersystemsforpatternrecognitionandmachinelearning.Therefore,Featureselectionhasattractedmuchattentionfrompatternrecognitionandmachinelearningsociety[1,2,3,4].TheaimofFSisto ndasubsetoforigi-nalfeaturesfromagivendatasetthroughremovingirrelevantand/orredundantfeatures[5].FScanimprovethepredictionperformanceofthepredictions,pro-videfasterandmorecost-e ectivepredictorsandgiveabetterinsightoftheunderlyingprocessthatgeneratedthedata[3,4].

Traditionalstatisticalmethodsforclusteringandclassi cationhavebeenex-tensivelyusedforfeatureselection.SupportVectorMachines(SVMs),proposedin1990sbyVapnik,havealreadybeenknownasane cienttoolfordiscover-inginformativeattributes[6].SVMsminimizetheupperboundoftrueerrorbymaximizingmarginwithwhichtheyseparatethedataintodistinctclasses.Intheframeworkofstatisticallearningtheory,theupperboundoferrorsofaclassi erdependsonitsVCdimensionandthenumberofthetrainingdata.FeaturesubsetselectionmayleadtoreducetheVCdimensionofthetrainedclassi er,andthenimprovestheclassi cationperformance.Alotofempiricalanalysisshowedintheliteraturessupportsthisargument.

J.Wangetal.(Eds.):ISNN2006,LNCS3971,pp.1373–1380,2006.

cSpringer-VerlagBerlinHeidelberg2006

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email