A classification algorithm for multi-classes based on SVM

Abstract

Support vector machine (SVM) constructs an optimal hyperplane utilizing a small set of vectors near boundary. The proximal SVM is an extremely simple procedure to generate linear and nonlinear classifier based on proximity to one of two parallel planes that are separated as far as possible. However, when the two-class are very unbalanced, the proximal SVM tends to fit better the class with more samples and has higher error in fewer samples. Further more, this draw back exists in K-category classification by using one-from-the-rest (OFR) separation for each class. To solve the problem, an improved SVM algorithm is presented in this paper. Experimental results show that the novel approach is prior to the proximal SVM.