Least squares algorithm for hard spaced support vector machines

• 1、Mathematics and Statistics School, Chongqing University, Chongqing 401331
• 2、Mathematics and Statistics School, Yili Normal University, Xinjiang 835000

Abstract：The fully linearly separable hard interval support vector machine (SVM) is a convex quadratic programming problem. Generally, SMO algorithm is used. If the sample size of training data is large, the training of SVM requires more computation and memory. In order to overcome this shortcoming, this paper transforms the training of support vector machine into the least square iterative solution. The method is to transform the hard interval support vector machine into the least square method model. The support vector machine is only related to the support vector, and has nothing to do with other sample points. The solution of the least square method is related to the number of samples. By continuously reducing the search range of support vector, the reduced range is used The least square solution is obtained by using the sample data set of the least square method model, and the convergence solution is the support vector machine solution. At the end of the paper, the least square method is used to solve the classification support vector machine of linear separable iris data set. The result is consistent with the convex quadratic programming solution of support vector machine, which shows that the least square method is feasible and effective.

Keywords： Hard spaced support vector machine least square method convex quadratic programming equations