Support vector machine for classification based on linear programming
XIE Hong1, LIU Heli1, WEI Jiangpin2( 1. Info. Eng. College, Shanghai Maritime Univ., Shanghai 200135, China; 2. Dept. of Computer Eng., Jiangsu College of Info. Tech., Wuxi Jiangsu 214061, China)
With the application of Support Vector Machine (SVM) on classification problem, the complexity of general norm control model in structure risk is analyzed, and two kinds of linear programming support vector machine are presented based on l1-norm and l∞-norm including linear SVM and nonlinear SVM. A numerical experiment is done for four kinds of SVMs which are three kinds of linear programming SVMs (including two kinds of proposed SVMs) and classic quadratic programming SVM. The experiment results show that in linear SVM case the parameters computed by three kinds of linear programming SVMs are near to theoretic value, and in nonlinear SVM case l1-norm-based SVM has least support vectors and better learning effect than the other SVMs.