A New Fast Twin Support Vector Regression

Twin support vector regression (TSVR) was proposed recently as a novel regression algorithm that determines a pair of -insensitive up- and down-bound functions by solving two related SVM-type problems, each of which is smaller than that in a classical SVR. However, it lack of complexity control and only implements empirical risk minimization principle. This paper proposed modified twin support vector regression that implements structural risk minimization principle by introducing the regularization term based on TSVR. The optimization problems are solved by successive overrelaxation technique. Experimental results on several datasets show good generalization performance and decreases computation time.