基于優(yōu)化的分類超平面支持向量算法研究.pdf

1、分類號(hào)密級(jí)UDC編號(hào)桂林電子科技大學(xué)碩士學(xué)位論文題目基于優(yōu)化的分類超平面支持向量算法研究基于優(yōu)化的分類超平面支持向量算法研究(英文英文)ResearchofAlgithmsaboutSupptVectOnTheClassificationHyperplaneBaseonOptimization研究生姓名:徐健指導(dǎo)教師姓名、職務(wù)指導(dǎo)教師姓名、職務(wù):陳光喜副教授申請(qǐng)學(xué)位門類:理學(xué)碩士學(xué)科、?？啤I(yè):應(yīng)用數(shù)學(xué)提交論文日期:2006年12月論

2、文答辯日期:2007年4月2007年4月6日AbstractKeyWdsABSTRACTTraditionalmethodsofMachineLearning(ML)supposethatsamplesetapproachtoinfinite.Usuallysamplesetisfiniteinpractice.StatisticalLearningThey(SLT)hasbeenproposedtosolvethisproblemup

3、tonow.SupptVectMachine(SVM)isamethodbasedonSLT.BecauseofstrictintheyprincipalgoodperfmanceinapplicationSVMhasalreadybeenahighlightinMLDataMining(DM).Themainwksinthispaperaretoimprovesomealgithmsdicussetheyindertoraisethe

4、speedofcalculationwhenSVMdealswiththelargesampleset.Thenumericalexperimentsapplicationsindicateefficiencyofthesealgithms.Thewksarestatedasfollow:IntheysystemthepaperproposesthatSVMcanbedividedintoalgebraicmethodgeometric

5、methodbyvirtueofhalvinglatestmethodmaximalspacingmethod.Thealgebraicmethodgeometricmethodareprovedtogainthesamehyperplane.Therelationbetweenquadraticprogrammingwithalgebraicmethodisexplained.Inviewofeffectofprojectionfin

6、dingthelatestpointsintwoconvexhullsisreplacedwithfindingtheminimumEuclideannmalinaconvexhull.Someconvergenceofalgithmsisanaylized.Inalgithmfirstlyanewalgithmcombiningmatrixtransfmationoffmulaissetuptogetsupptvect.Thensom

7、eimprovementsaregainedbyanalyzingSequentialMinimalOptimization(SMO).FexamplethresholdsearchinginSMOisimprovedacacheissetuptocontrolcomputationalvalues.Apreprocessingmethodisbroughtfward.Themethodscommonspeciespointsdiffe

8、rentspeciespointsisolatedpointssoastopointsexpressingtheacterofsamplesetrepeatedly.Finallythearticledesignsaniterativealgithmfgeometricmethod.Inthenumericalexperimentsapplicationstheresultsofthenumericalexperimentverifyt

9、heefficiencyofalgithms.SVMthepreprocessingmethodareappliedinmedicinediagnosisfAmericanWisconsinBreastCancerDatabase.Theresultindicatesthatthepreprocessingmethodnotonlyreducestrainingtimebutalsomaintainsclassificationaccu