A Novel ART-based Multi-class Classification Method for SVM

Based on the principle of binary classification of support vector machine (SVM),an improved multi-class classification method is developed,which combines the adaptive resonance theory with SVM.The proposed approach improves the one-against-one classifying algorithm of traditional SVMs.In the decision-making process of the results of each binary classifier,the voting principle is not adopted;instead the adaptive resonance theory is used to fuse the output of each binary classifier.Thus this method avoids the existence of fusing errors when the binary classifier outputs approach zero,and overcomes the problem of refusing to fuse when the algorithm gets the same votes.The algorithm has been applied to glass-classification.Simulation experiments prove that the classification results are more accurate.