支持向量机利用接近边界的少数向量来构造一个最优分类面。然而当两类中的样本数量差别悬殊时,PSVM算法则会过度拟合样本量大的那一类,而对样本量很小的那一类的错分率相当高。为解决此问题,本文提出了一种改进的支持向量机算于拟牛顿法的大类别分类算法。同时,这个问题也是大类别分类问题所采用的留一法面临的问题,在DFP-PSVM的基础上,提出了基于拟牛顿法的大类别分类算法。通过仿真实验证实了此算法在精度上优于PSVM算法。
Support vector machine constructs an optimal hyperplane utilizing a small set of vectors near boundary. However, when the two-class problem samples are very unbalanced, PSVM tends to fit better the class with more samples and has high error in fewer samples. To solve the problem, an improved SVM algorithm, DFP-PSVM, is presented in this paper. Furthermore, this drawback exists in one-from-the-rest approach to multi-classes. A multi-class classification algorithm using quasiNewton is proposed based on DFP-PSVM. Simulated examples show that the novel algorithm is prior to the plain PSVM.