A regression-type support vector machine for k-class problem

In this paper, a new method for multiclass classification is proposed, named regression based support vector machine (RBSVM). Rest-versus-1-versus-1 strategy is adopted in RBSVM. When constructing the classification hyperplane, a mixed regression and classification formulation is used. The training set is divided into positive class, negative class and rest class, the regression hyperplane is required to pass through the rest class and it can classify the positive and negative class correctly as far as possible simultaneously. The proposed method can not only improve the classification accuracy but also reduce the computational complexity compared with some currently existed multiclass classification methods. The numerical results show the effectiveness of our methods.