A Novel Decision-making Method for Multi-class SVMs and Its Application

This paper presents several existing multi-class classifiers and compares their advantages and disadvantages. Then,a decision-making method for multi-class SVMs is proposed to recombine the outputs of several binary classifiers.In the strategy,a novel decision function is defined,and its value is the traditional voting values multiplied by the corresponding weights of different classifiers.The presented multi-class SVM is of better classification ability and can solve the unclassifiable region problems in traditional max-wins-voting(MWV) strategy.Lastly,the novel method is used in fault diagnosis as the key technology,and the actual diagnosing results demonstrate the superiority of the improved strategy over the traditional one.