Support Vector Regression with Automatic Margin Control

Support vector regression (SVR) is a typical regression method, and has been successfully applied in many practical problems such as financial engineering. However, the conventional SVR depends mainly on the size of ε-insensitive margin which is unsuitable especially when samples are volatile and noisy. In this paper, we proposed a novel regression algorithm, termed as v-support vector regression with automatic margin control (AMC-v-SVR), to tackle this problem. AMC-v-SVR seeks the regressor and its up- and down- margins simultaneously by solving a single quadratic programming problem. The proposed regression algorithms have the advantage when the margin is not fixed and asymmetrical. Experimental results show the feasibility and effectiveness of the proposed method.