用于多分类问题的最小二乘支持向量分类―回归机.docx

用于多分类问题的最小二乘支持向量分类―回归机文章编号:10019081(2013)07189404doi:10.11772/j.issn.10019081.2013.07.1894摘要：基于支持向量机(SVM)的三分类方法是处理多分类问题的一类方法。提出了最小二乘支持向量分类回归机(LSSVCR)算法通过最小二乘目标函数充分考虑所有样本点对分类的影响使得训练集中即使有个别样本点被标错类别对分类结果也不会产生太大的影响从而提高分类的准确性。该方法能够提高分类的准确率和分类速度同时算法对于不同类别间样本数目差异较大的情况也有很好的分类效果。数值实验结果表明所提算法是可行的且与已有的三分类算法相比在分类准确性上平均提高了2.57%在运算速度上也有了较大的提高。关键词：多分类问题；三分类问题；最小二乘支持向量机；分类回归机；一对一对多方法中图分类号：TP181文献标志码:A英文标题Leastsquaresupportvectorclassificationregressionmachineformulticlassificationproblems英文作者名ZHAIJiaHUYiqing*XUEr英文地址(SchoolofMathematicsandPhysicsUniversityofScienceandTechnologyBeijingBeijing100083China英文摘要)Abstract：TriclassclassificationmethodbasedonSupportVectorMachine(SVM)isakindofmethodforsolvingmulticlassclassificationproblems.LeastSquareSupportVectorClassificationRegression(LSSVCR)wasproposedwhichconsideredtheeffectsofallthesamplepointsbyusingleastsquaresobjectivefunction.Eveniftherewerewronglymarkedsamplepointsinthetrainingsettheresultwouldnotbeaffectedlargelybythem.LSSVCRwasmoreaccurateandfasteranditwasefficientfortheproblemsthattherearelargedifferencesamongthenumberofsamplepointsindifferentclasses.Thenumericalexperimentsshowthattheproposedmethodraisestheaccuracyby2.57%onaveragecomparedtotheexistingtriclassificationmethods.TriclassclassificationmethodbasedonSupportVectorMachine(SVM)isakindofmethodsolvingmulticlassclassificationproblems.LeastSquareSupportVectorClassificationRegression(LSSVCR)wasproposedwhichconsideredtheeffectsofallthesamplepointsbyusingleastsquaresobjectivefunction.Eveniftherewerewronglymarkedsamplepointsinthetrainingsettheresultwouldnotbeaffectedlargelybythem.LSSVCRwasmoreaccurateandfasteranditwasefficientfortheproblemswhichhadlargedifferencesinthenumber