A full smooth semi-support vector machine based on the cubic spline function

Abstract

The non-smooth problem for the semi-supervised support vector machine optimization model is studied. Since the objective function of the unstrained semi-supervised vector machine model is a non-smooth function. Most fast optimization algorithms can not be applied to solve the semi-supervised vector machine model. We propose a full smooth cubic spline function to approximate the symmetric hinge loss function. The Broyden-Fletcher-Goldfarb-Shanno(BFGS) algorithm is used to solve the new model. The experimental results show that the new model has a better classification performance.