Locality Preserving Semi-Supervised Support Vector Machine

Manifold regularization, which learns from a limited number of labeled samples and a large number of unlabeled samples, is a powerful semi-supervised classifier with a solid theoretical foundation. However, manifold regularization has the tendency to misclassify data near the boundaries of different classes during the classification process. In this paper, we propose a novel classification method called locality preserving semi-supervised support vector machine (LPSSVM) with an extended manifold regularization framework based on within-class locality preserving scatter. LPSSVM is good at exploring the underlying discriminative information as well as the local geometry of the samples as much as possible rather than merely relying on the smoothness information regarding manifold regularization. Meanwhile, benefiting from the geodesic distance metric, LPSSVM can more effectively reflect the true local geometry of data instances in the manifold space, which further strengths its accuracy in reality. The extensive comparisons with respect to LPSSVM and several state-of-the-art approaches were carried out on both artificial and real-word data sets. These experimental studies demonstrate the advantages as well as the superiority of our proposed method.