一种新的二叉树多类支持向量机算法  

An Improved Multiclass Support Vector Machines Based on Binary Tree

作　　者：唐发明[1] 王仲东[1] 陈绵云[1] 

机构地区：[1]华中科技大学控制科学与工程系,武汉430074

出　　处：《计算机工程与应用》2005年第7期24-26,共3页Computer Engineering and Applications

基　　金：国家自然科学基金;国家部委科技预研基金项目资助

摘　　要：采用二叉树结构对多个二值支持向量机(SVM)子分类器组合,可实现多类问题的分类,并且还可克服传统多类SVM算法存在的不可分区域的情况。针对现有二叉树多类SVM方法未采用有效的二叉树生成算法,该文采用聚类分析中的类距离思想,提出了一种新的基于二叉树的多类SVM分类方法。实验结果表明,新算法具有较高的推广性能。The method constructing and combining several binary SVMs(Support Vector Machines) with a binary tree can solve multiclass problems,and resolve the unclassifiable regions that exist in the conventional multiclass SVMs.Since some existing methods based on binary tree don't use an effective constructing algorithm of binary tree,this paper proposes an improved multiclass SVM based on binary tree using class distance of clustering.Numerical experiments on large problems demonstrate the effectiveness of the method over conventional methods.

关 键 词：多类支持向量机 聚类 二叉树 多类分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]