A Novel Smooth Support Vector Regression Based On Taylor Formula

作者：Ren, Bin^*; Cheng, LiangLun

来源：Information-An International Interdisciplinary Journal, 2012, 15(1): 23-32.

• epsilon-insensitive loss function
• Taylor formula
• smooth approximation
• support vector regression

Conventional E-Support Vector Regression is formulated as a constrained minimization problem, i.e., a convex quadratic programming problem. Lee et al. used the smoothing techniques to replace the square of E-insensitive loss function and named the reformulation Smooth E-Support Vector Regressiop. Thus E-SVR can be solved as a smooth unconstrained minimization problem by the Newton-Armijo algorithm directly. This paper takes Taylor formula to generate a new polynomial smooth function vertical bar x vertical bar(2)(epsilon). in epsilon-insensitive support vector regression. Theoretical analysis shows that T-epsilon(2) -function is better than p(epsilon)(2)-function and S-epsilon(2)-function in properties, and the approximation accuracy of the proposed smoothing function is three order of higher than that of classical p(epsilon)(2)-function. The Numerical results show the efficiency of the new approach.

• 出版日期2012-1
• 单位东莞理工学院; 广东工业大学