Enhancing ε-support Vector Regression with Gradient Information
Xiao Tao Zhou · Acta Automatica Sinica · 2014
Traditional methods constructing of e-support vector regression(e-SVR) do not consider the gradients of the true function but only deal with the exact responses at the samples. If the gradient information is available easily and cheaply, it should be used to enhance the model. The existing research on constructing of e-SVR with gradient information starts from the perspective of Taylor expansion, and simply inserts the additional objective values in the neighborhood of the sampled points into the corresponding terms of a Taylor expansion. In this paper, the gradient-enhanced e-support vector regression(GESVR) is developed with a direct formulation by incorporating the gradient information into the kernel matrix. The efficiency of this technique is verified by analytical function fitting and the actuarial data in the life table. The results show that the GESVR provides more reliable prediction results than e-SVR alone.