Multi-class SVM classifiers fusion based on evidence combination
Deqiang Han, Chongzhao Han, Yi Yang · 2007
An approach to implementing multiple classifiers fusion based on evidence combination is proposed in this paper. The member classifiers are designed based on the multi-class SVM. In common, the output of multi-class SVM classifier is just the label of class. Based on the confusion matrix and the class-wise performance, we propose a novel approach to generating the mass functions, which can reduce the computational complexity of evidence combination. Independent member classifiers are trained based on heterogeneous features. And then the fusion of multiple classifiers fusion can be implemented based on Dempster rule of combination. Experimental results provided show the efficacy and rationality of the novel approach proposed.