A Maximum Class Distance Support Vector Machine

A maximum class distance based support vector machine classification algorithm (MCDSVM) using Fisher linear discriminant analysis (FLDA) is presented in this paper. The algorithm can maximize the margin between the separating hyperplane and the distance between the samples of two classes. The direction of separating hyperplane can be consistent with the distribution of samples and the algorithm can achieve higher classification accuracy. This algorithm can also overcome the over-fitting of SVM resulting from outliers, as well as the problem that the hyperplane doesn´t adapt to the distribution of samples. The principle and realization of the algorithm are addressed in detail in this paper and the classification performance is also analyzed in theory. Finally, a simulation demonstrates the efficiency of this new algorithm.