A new proximal support vector machine for semi-supervised classification

Proximal support vector machine (PSVM) is proposed instead of SVM, which leads to an extremely fast and simple algorithm by solving a single system of linear equations. However, sometimes the result of PSVM is not accurate especially when the training set is small and inadequate. In this paper, a new PSVM for semi-supervised classification (pS(3)VM) is introduced to construct the classifier using both the training set and the working set. pS3VM utilizes the additional information of the unlabeled samples from the working set and acquires better classification performance than PSVM when insufficient training information is available. The proposed pS3VM model is no longer a quadratic programming (QP) problem, so a new algorithm has been derived. Our experimental results show that pS3VM yields better performance.