一种基于FCOWA-ER的SVM多分类方法

A multi-class SVM based on FCOWA-ER

支持向量机(SVM)在处理多分类问题时,需要综合利用多个二分类SVM,以获得多分类判决结果.传统多分类拓展方法使用的是SVM的硬输出,在一定程度上造成了信息的丢失.为了更加充分地利用信息,提出一种基于证据推理-多属性决策方法的SVM多分类算法,将多分类问题视为一个多属性决策问题,使用证据推理-模糊谨慎有序加权平均方法 (FCOWA-ER)实现SVM的多分类判决.实验结果表明,所提出方法可以获得更高的分类精度.

Multiple bi-class SVMs are used together to obtain the final decision when the support vector machine（SVM） is applied to multi-class classification problems. The conventional methods of applying the SVM to multiple classification tasks are all based on the hard output of SVM, which can bring the loss of information to some extent. Therefore, a multi-class SVM based on an evidential reasoning based multiple attribute decision approach is proposed to use more information. The multi-class classification problem is modelled as a multi-criteria decision making problem. Then a fuzzycautious OWA（ordered weighted averaging） approach with evidential reasoning（FCOWA-ER） is used to implement multiclass classification and obtain the final decision. The simulation results show that the method proposed has better accuracy compared with conventional methods.