A multi-classification algorithm based on support vectors

Abstract

In the fault classification process, a flexible SVM classification algorithm is proposed to solve the unreasonable condition that the number of muti-classification decision boundary is stationary when using the traditional support vector machine(SVM). The algorithm is based on support vector data description(SVDD) hypersphere determine the sample distribution characteristics similar class of fusion as a new class, guaranted to produce classifications which are easy to distinguish. Training multi hyperspheres between the new classes and SVM decision boundary within the new class. Using one-to-one vote to choose. Experiments show that this algorithm has a better classification performance, and can reduce training time and determine time which can be well applied to fault classification.