Locality Sensitive Proximal Classifier with Consistency for Small Sample Size Problem

The proximal classifier with consistency (PCC) is an improvement of generalized eigenvalue proximal support vector machine (GEPSVM), ensuring consistency ignored in GEPSVM. However, similar to many other machine learning methods, PCC uses only the global information and the eigenvalue problem need to be solved, which can not classify small sample size (SSS) problem effectively. By exploiting both global and local information, we propose a novel binary classifier named locality sensitive proximal classifier with consistency (LSPCC). Our LSPCC determines two proximal hyperplanes by solving two small eigenvalue problems. This makes our LSPCC is able to deal with SSS problem well. Experimental results on several standard small sample size datasets have shown the superiority of our approach.