关键词: 多类分类问题, 广义特征值, 支持向量机

Abstract: In this paper, it proposes a new multi-class classification algorithm. The task is to find [M] non-parallel hyperplanes. The [m-th(m=1,2,?,M)] hyperplane is constructed such that the inputs of the [m-th] class are close to the hyperplane and the rest inputs are far away from it as far as possible. Based on this idea, for finding the [m-th] hyperplane, the optimization model can be transformed into a generalized eigenvalue problem. The method is easily implemented by a single Matlab command that solves the classical generalized eigenvalue problem. The preliminary numerical experiments show that the method is competitive with several classical multi-class classification algorithms based on support vector machine.

Key words: multi-class classification problem, generalized eigenvalue, support vector machine