用于非平衡样本分类的近似支持向量机

Proximal Support Vector Machines for Samples with Unbalanced Classification

针对标准的近似支持向量机(PSVM)没有考虑样本分布不平衡的问题,提出一种改进的 PSVM 算法(MPSVM).根据训练样本数量的不平衡对正负样本集分别分配不同的惩罚因子,并将原始优化问题中的惩罚因子由数值变更为一个对角阵.最后推导出线性和非线性 MPSVM 的决策函数,并将其与 PSVM、非平衡的 SVM 的运算机理和性能进行比较.实验结果表明,MPSVM 的性能优于 PSVM,与非平衡 SVM 方法相比效率更高.

Aiming at the problem that unbalanced data classification is disregarded in the standard Proximal Support Vector Machines （PSVM）, a modified PSVM algorithm is presented, namely MPSVM . The different penalty factors are assigned to the positive and negative training sets according to the unbalanced population . The penalty values are transformed into a diagonal matrix. Then the decision functions for the linear and nonlinear MPSVM are achieved. Finally, the comparisons of algorithmic principle and performance are drawn. The experimental results show that MPSVM has a better generalization performance than PSVM and higher efficiency than the unbalanced SVM.