Weighted Linear Loss Support Vector Machine for Large Scale Problems

In this paper, instead of using the Hinge loss in standard support vector machine, we introduce a weighted linear loss function and propose a weighted linear loss support vector machine (WLSVM) for large scale problems. The main characteristics of our WLSVM are: (1) by adding the weights on linear loss, the training points in the different positions are proposed to give different penalties, avoiding over-fitting to a certain extent and yielding better generalization ability than linear loss. (2) by only computing very simple mathematical expressions to obtain the separating hyperplane, the large scale problems can be easy dealt. All experiments on synthetic and real data sets show that our WLSVM is comparable to SVM and LS-SVM in classification accuracy but with needs computation time, especially for large scale problems.