基于Huffman树的多类SVM方法  

Multiclass SVM method based on Huffman tree

作　　者：刘洋[1] 张秋余[2] 

机构地区：[1]渤海大学信息科学与工程学院,辽宁锦州121000 [2]兰州理工大学计算机与通信学院,甘肃兰州730050

出　　处：《计算机工程与设计》2008年第7期1792-1793,1840,共3页Computer Engineering and Design

基　　金：甘肃省科技攻关计划基金项目(2GS047-A52-002-03)

摘　　要：提出了一种基于Huffman树的支持向量机多类分类方法。二叉树方法是一种常用的多类分类方法,它的关键问题在于如何构造合理的结构以获得较高的推广能力。为解决该问题,按照Huffman树的构造过程自下向上地构造二叉树,使易于分割的类处于上层结点。实验结果表明,该方法与One-vs-One和DAGSVM方法的分类效果相当。A support vector machine multiclass classification method based on Huffman tree is put forward. The binary tree method of multiclass classification is an ordinary method, the key issue of which is how to construct a reasonable structure to maintain high gene- ralization ability. To solve the problem, a binary tree is constructed from down to up according to the construction process of Huffman tree, making the class easy to separate lies in the upper node. Numerical experiments show that the method proposed is comparative to the one-vs-one and DAGSVM methods which have better classification accuracy.

关 键 词：机器学习 支持向量机 多类分类 二叉树 哈夫曼树 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]