v支持向量機中參數(shù)選取的兩種模型

1、大連理工大學(xué)碩士學(xué)位論文v支持向量機中參數(shù)選取的兩種模型姓名：張小路申請學(xué)位級別：碩士專業(yè)：運籌學(xué)與控制論指導(dǎo)教師：夏尊銓20080605大連理工大學(xué)碩士學(xué)位論文TwomodelsofParameterSelectioninvSupport昵ctorMachineAbstractWiththedevelopmentofmoderninformationaltechnologyalargenumberofdatabaseshaveappe

2、aredConsequentlyhowtofindusefulinformationfromsetsofdatabecomesaproblemneedtobesolvedimminentlyDataminingtechnologycomesintobeinginthisbackgroundSupportvectormachineisanewtechnologyofdatamining，andithasbeensuccessfullyap

3、pliedtomanyfieldsSupportvectormachine(SVM)isamachinelearningmethodthatdevelopedonthebasisofstatisticallearningtheoryonthemid1990sItcontainsthelargestmarginbyperplane，Mercerkernal，convexquadraticprogramming，slackvariables

4、andsuchtechnologiesUntilnowithasgotthebestpropertiesinsomechallengeableapplicationsT婦papermainlystudiestheoptimalmethodsforparameterselectioninz，一supportvectormachine(v—SVM)The∥一supportvectormachineisanewclassofsupportve

5、ctoralgorithms睢SVMWasproposedbyincorporatingachangefromCintheoriginalSVMalgorithmwith∥whichpossessesthefollowingintuitivemeaning：z／isbothanupperboundonthefractionofmarginerrorsandalowerboundonthefractionofsupportvectorsT

6、heparameterselectioninsupportvectormachineisanimportantresearchtopic，differentparametersresultindifferentgeneralizationThestudyofparameterselectionin∥一SVMisfewerTheparameterselectionmodelfor∥in∥SVMispresentedandthenumeri

7、calexperimentisgiveninthispaperAtfirst，amodelforselectingoptimalparameter∥inpSVMispresentedinthispaper，whichbasesonmathematicalprogrammingwithequilibriumconstraintsThemodelisanonhnearproblemwhichhassmoothedobjectivefunct

8、ionandwithcomplementaryconstrainsNumericalexperimentalresultsshowthatthismodeliseffectiveinchoosingtheparameter∥Besideswediscussthez，SVMwhenthedatasetsareunbalancedAnimproved2∥SVMispresentedinthispaperthismodel’Smeritist

9、heparametereasytoselectNumericalexperimentalresultsshowthatthismodeliseffectivewhensolvingunbalancedquestionKeyWords：SupportVectorMachine(SVM)；∥一SupportVectorMachine(u—SVM)；ParameterSelection；MathematicalProgrammingUnbal