Locality Preserving One-Class Support Vector Machine

One-class support vector machine (OCSVM) tries to find a hyperplane to distinguish normal data from all other possible outliers or abnormal data. However, only support vectors determine the hyperplane. In this paper, we propose a novel data description method called locality preserving one-class support vector machine (LPOCSVM). It takes the intrinsic manifold structure of data into full consideration. In the paper, we discuss the linear and nonlinear case of LPOCSVM, and detail how to tackle the singularity of the locality preserving scatter matrix. Experimental results on several toy and benchmark datasets indicate the effectiveness and advantage of LPOCSVM by comparing it with OCSVM.