一种快速最小二乘支持向量机分类算法.pdf

168 , ( ) 计算机工程与应用 ComputerEngineeringandApplications 一种快速最小二乘支持向量机分类算法 孔 锐,张 冰 , KONGRuiZHANGBing 暨南大学 珠海学院 计算机科学系,广东 珠海 519070 , , , , DepartmentofComputerScienceofZhuhaiCollegeJinanUniversityZhuhaiGuangdong519070China : E-mailtkongrui@ , KONGRuiZHANGBing.Classificationalgorithm offastleastsquaressupportvectormachine.ComputerEngineeringand , , ( ): Application168-170. : ( ) , AbstractLeastSquaresSupportVectorMachinesLS-SVM acquiretheoptimalsolutionbysolvingasetoflinearequationsin- , steadofsolvingaconvexquadraticprogrammingproblemButthesolutionsinlosesparsityproperty.Whenthetrainsetofsample , pointsisbiggerthecostofcomputationbecomesgreat.ThepaperpresentsanewalgorithmofFastLeastSquaresSupportVector ( ) , , Machines FLS-SVM.Asthesamegeneralizationabilityespeciallywhenthetrainsetofsamplepointsisbiggerthetrainspeed ofthenewalgorithm isfasterthanthatoforiginalLS-SVM algorithm.Thenewalgorithm firstselectsthesamplesasreduced trainingsetwhichhavebiggersupportvaluefromtotaltrainingset.ThenittrainsLS-SVM toacquireoptimalsolutionbyusing theselectedsamplesinreducedtrainingset.Theresultsofexperimentverifythatthenewalgorithmnotonlyacquiresthesame , generalizationabilitywiththatoftheoriginalalgorithmsbutalsoisfasterthanthatoftheoriginalalgorithms. : ; ( ); ; ( ) Keywords sparsitypropertyLeastSquaresSupportVectorMachinesLSVM kernelfunctionSupportVectorMachinesSVM 摘 要:最小二乘支持向量机不需要求解凸二次规划问题,通过求解一组线性方程而获得最优分类面,但是,最小二乘支持向量机 失去了解的稀疏性,当训练样本数量较大时,算法的计算量非常大。提出了一种快速最小二乘支持向量机