A Maximum Class Distance Robust Support Vector Machine Classification Algorithm

A maximum class distance robust support vector machine (MCDRSVM) is presented in this paper.Inspired from the principles of linear discriminate analysis(LDA) and support vector machine(SVM), the method adapts the robust cost function and tries to find a optimal separable hyperplane by maximizing the class scatter distance, which reduces influences of outliers and greatly improves the performance of the proposed algorithm. To deal with the singularity of the class matrix resulted with the small size sample,kernel principal component analysis (KPCA) is applied to transform samples to lower dimension, then the hyperplane can be achieved by solving MCDRSVM optimization problem. The simulations demonstrate the efficiencies of the proposed algorithm.