Improved weighted support vector regression
LI Zhonghao, WANG Yu (School of Management, Dalian Univ. of Tech., Dalian Liaoning 116024, China)
The disadvantage and insufficiency of the existing methods of support vector regression (SVM) are analyzed. An improved weighted support vector regression and its Wolfe Dual is presented. The convex functions and its reformation depending on the various types of the kernel function are introduced to decrease the limitation of the kernel function. It is helpful to get more effective in searching flexible regression function. The generalization of the support vector regression model, the optimization of the generalization capacity, and the training speed are discussed.