A new proximal support vector machine for semi-supervised classification

Abstract

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. PS 3 VM 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 PS 3 VM model is no longer a quadratic programming (QP) problem, so a new algorithm has been derived. Our experimental results show that PS 3 VM yields better performance. © Springer-Verlag Berlin Heidelberg 2006.