Multi-class support vector machines with binary tree architecture (SVM-BTA) have the fastest decision-making speed in the existing multi-class SVMs. But SVM-BTA usually has bad classification capability. According to internal characteristics of feature samples, this paper uses resemblance coefficient method to construct automatically binary tree to incorporate multiple binary SVMs. The multi-class SVMs with constructed binary tree have good classification capability and fast decision-making speed. Experimental results of yeast protein localization site prediction and radar emitter signal recognition show that the introduced multi-class SVMs with binary tree architectures are superior to several popular multi-class SVMs including one-against-all, one-against-one, directed acyclic graph, bottom-up binary tree and several classification methods in the recent literatures.
展开▼
