基于SVM概率输出与证据理论的多分类方法

Multi-class Classification Method Based on SVM Probability Output and Evidence Theory

单一技术无法有效解决多类分类问题。为此,提出一种基于一对多支持向量机(SVM)的基本概率分配输出方法,并与置信最大熵模型的D-S证据组合方法结合,给出基于SVM概率输出和证据理论的多分类模型。在3种UCI标准数据集上的仿真结果表明,该方法的分类精度优于传统的一对多和一对一硬输出方法,是一种有效的多类分类方法。

One-technology do not solve multi-class classification problem,on the basis of this,a basic probability output distribution method based on One-Against-All（OAA） Support Vector Machine（SVM） is proposed,a multi-class model based on Support Vector Machine（SVM） probability output and evidence theory is put forward by integrating one-against-all multi-class SVM with max-entropy D-S theory,.Simulations results on three datasets of UCI repository show that the method has higher classification precision than hard output method OAA and OAO.