一种新的SVM决策树  

A new SVM decision tree

作　　者：孙昌儿[1] 刘秉瀚[1] 

机构地区：[1]福州大学数学与计算机科学学院,福建福州350002

出　　处：《福州大学学报（自然科学版）》2007年第3期361-364,共4页Journal of Fuzhou University(Natural Science Edition)

基　　金：福建省自然科学基金资助项目(A0610013);国家自然科学基金资助项目(60372027;60675058)

摘　　要：SVM在小训练样本、高维情况下具有很好的泛化性能,但它不适用于多类分类.本文分析基本的SVM和多类SVM分类器,重点讨论了SVM决策树,提出了一种结点分类器类集合划分方案来构造SVM决策树.实验结果表明,以这种方法构造的SVM决策树分类器分类性能较好.SVM has good generalization performance when the number of training samples is very small and the dimension of feature space is very high. But it is not suitable for multi -class classification. This paper analyzes the basic SVM and the SVM classifier multi - class classification, especially about the SVM decision tree, then proposes a method for partition of the set of classes on each node classifier to build up SVM decision tree. The results of experiment demonstrate that the SVM decision tree built up by this method has a good classification Performance.

关 键 词：支持向量机 SVM决策树 分类器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]