A Robust Proximal Support Vector Machines for Classification

Proximal support vector machines (PSVMs) are support vector machines (SVMs) version which involve equality instead of inequality constraints assigning data points to one of two classes, and have a significant speedup in comparison with traditional SVM-type classifiers. However, PSVMs are very sensitive to outliers or noises because of overfitting problem. In this paper, a robust proximal support vector machines (RPSVMs) is proposed to deal with the problem of overfitting in PSVMs for two-class classification. In experiments, two methods are used to give membership values to training data points according to their relative importance in the training set. Results show that the proposed RPSVMs actually reduce the effect of outliers and yield higher classification rate than PSVMs do