支持向量机方法中加权后验概率建模方法

Weighted posterior probability output for support vector machines

为解决传统支持向量机方法不提供概率输出的问题,在支持向量机多类分类问题输出概率建模中,提出了加权后验概率建模方法。该方法在对多个两类支持向量机分类器的输出概率进行组合时,充分考虑了各个两类支持向量机分类器的差异。依据Bayesian理论,采用后验概率作为各个两类支持向量机分类器的权系数。实验结果表明,与投票法及Pairwise Coupling方法相比,加权后验概率方法具有较低的分类错误率,不仅具有较好的分类性能,而且得到的后验概率具有较好的概率分布形态。该方法有效地解决了实际多类分类问题中支持向量机的概率建模问题。

A weighted posterior probability method is presented to calculate the probability outputs of support vector machines （SVMs） for multi class cases. The differences and weights for combination of the probabilty output among these two-class classifiers calculated from the posterior probability are given based on the Bayesian theory. Tests show that the weighted posterior probability method has less classification errors, better classification ability, and a better probability distribution of the posterior probability than the voting method or the Pairwise Coupling method. This method effectively provides probability outputs of SVMs in the multi-class case.