Applying Discriminant Functions with One-Class SVMS for Multi-Class Classification

Abstract

Early SVM-based multi-class classification algorithms work by splitting the original problem into a set of two-class sub-problems. The time and space required by these algorithms are very much demanding. We present in this paper a hybrid method that integrates several one-class SVMs with discriminant functions to solve the multi-class classification problem. Several discriminant functions, including similarity measure, distance measure, and Z-score measure, have been applied in this research. The proposed method has low time and space complexities. Experimental results show that our method compares favorably with SVDD-based multi-class classification algorithms on several real datasets from UCI and Statlog.