关键词: 回归问题, 支持向量回归机, 双支持向量回归机, 最小二乘双支持向量回归机

Abstract: In this paper Least Square Twin Support Vector Regression（LSTSVR） is proposed, which is formulated via the idea of Twin Support Vector Regression（TSVR）. LSTSVR breaks the idea which [ε]-band is constructed by two parallel hyperplanes in traditional Support Vector Regression（SVR）. Actually, LSTVR employes two non-parallel hyperplanes to construct the [ε]-band, in which each hyperplane determinates a half [ε]-bond, and obtain the final regression. So the regression function fits the distribution of dataset and the algorithm has better generalization ability. In addition, in LSTSVR, the main computing cost is to solve two smaller systems of linear equations, so the computational complexity is low. The experimental results indicate that the proposed algorithm has certain advantage in generalization ability and computational efficiency.

Key words: regression problem, support vector regression, twin support vector regression, least square twin support vector regression