Use of Multi-category Proximal SVM for Data Set Reduction

Abstract. We present a tutorial introduction to Support Vector Machines (SVM) and try to show using intuitive arguments as to why SVM’s tend to perform so well on a variety of challenging problems. We then discuss the quadratic optimization problem that arises as a result of the SVM formulation. We talk about a few computationally cheaper alternative formulations that have been developed in recent years. We go on to describe the Multi-category Proximal Support Vector Machines (MPSVM) in more detail. We propose a method for data set reduction by effective use of MPSVM. The linear MPSVM Formulation is used in an iterative manner to identify the outliers in the data set and eliminate them. A k-Nearest Neighbor (k-NN) classifier is able to classify points using this reduced data set without significant loss of accuracy. We also present geometrically motivated arguments to justify our approach. Experiments on a few publicly available OCR data sets validate our claims.