Unit Model of Binary SVM with DS Output and its Application in Multi-class SVM

A unit model of binary SVM with DS output is set up by using sigmoid function which maps the relative algebraic distance between the sample and the bound of each class into the BPA of DS evidence theory. As its application, a new multi-class SVM is designed, which is named mini-structure multi-class SVM because it is simpler than other multi-class SVM based on binary SVM. The discernment frame of the mini-structure model is partitioned according to the algorithm of modifying evidence sources, one of solutions to the combination of conflict evidences. Thus, the problem of the combination of conflict evidences is avoided. The simulation results show the excellent effect of the ways in this paper.