A new smooth support vector machine for classification

This paper studied smoothing support vector machine (SVM) for classification with a new class of smoothing functions. A new model, 3rd-order polynomial smooth support vector machine (3SSVM), was proposed, and its global convergence was established. Because of 3rd-order differentiability of the objective function of the unconstrained minimization problem, a fast Newton-Armijo algorithm can be used to solve 3SSVM. It has been shown that the sequence generated by this algorithm converges globally to the unique solution of the original classification problem with a smaller upper bound than that of Polynomial Smooth SVM (PSSVM). Numerical experiments were carried out to evaluate 3SSVM and the results confirmed the advantage of 3SSVM over PSSVM.