资源描述：
《A New Approach to Symbolic Classification Rule Extraction Based on SVM》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、ANewApproachtoSymbolicClassificationRuleExtractionBasedonSVMDexianZhang1,TiejunYang1,ZiqiangWang1,andYanfengFan21SchoolofInformationScienceandEngineering,HenanUniversityofTechnology,ZhengZhou450052,P.R.C(China)zdx@haut.edu.cn2ComputerCollege,NorthwesternPolytechnicalUniversity,X
2、i’an710072,P.R.C(China)Abstract.Therestillexisttwokeyproblemsrequiredtobesolvedintheclassificationruleextraction,i.e.howtoselectattributesanddis-cretizecontinuousattributeseffectively.Thelackofefficientheuristicinformationisthefundamentalreasonthataffectstheperformanceofcurrentlyuse
3、dapproaches.Inthispaper,anewmeasurefordetermin-ingtheimportanceleveloftheattributesbasedonthetrainedSVMisproposed,whichissuitableforbothcontinuousattributesanddis-creteattributes.Basedonthisnewmeasure,anewapproachforruleex-tractionfromtrainedSVMandclassificationproblemswithconti
4、nuousattributesisproposed.Theperformanceofthenewapproachisdemon-stratedbyseveralcomputingcases.Theexperimentalresultsprovethattheapproachproposedcanimprovethevalidityoftheextractedrulesremarkablycomparedwithotherruleextractingapproaches,especiallyforthecomplicatedclassificationp
5、roblems.1IntroductionOneofthedataminingproblemsisclassification.Classificationplaysaveryim-portantroleinmanyfieldsofapplications.Classificationistheprocessoffind-ingthecommonpropertiesamongdifferentpatternsandclassifyingthemintoclasses.Theresultsareoftenexpressedintheformofsymbolicru
6、les-theclassifi-cationrules.Byapplyingtherules,patternscanbeeasilyclassifiedintodifferentclassestheybelongto.Theclassificationruleextractionhasbecomeanimportantaspectofdatamining.Theexistingapproachesforextractingtheclassificationrulescanberoughlyclassifiedintotwocategories,datadrive
7、napproachesandmodeldrivenapproaches.Themaincharacteristicofthedatadrivenapproachesistoextractthesymbolicrulescompletelybasedonthetreatmentwiththesam-pledata.Themaincharacteristicofthemodeldrivenapproachesistoestablishamodelatfirstthroughthesampleset,andthenextractrulesbasedonthe
8、relationbetweeninputsandoutputsrepresentedbythemodel.T
