改进的基于平衡二叉决策树的SVM多分类算法  

An Improved Support Vector Machines with Balanced Binary Decision Tree for Multi-class Classification

作　　者：林志杰[1] 余春艳[1] 

机构地区：[1]福州大学数学与计算机科学学院,福州350108

出　　处：《小型微型计算机系统》2014年第5期1128-1132,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60805042)资助;福建省自然科学基金项目(2010J01329;2011J05150;2012J01262)资助;福建省重大产学合作项目(2010H6012;2011H6014)资助

摘　　要：基于平衡二叉决策树的支持向量机多分类算法的难点在于构造一棵有效的平衡二叉决策树.为构造有效的平衡二叉决策树,提出一种新的类间距离度量方案,进一步,引入类间可分性以及分离因子的概念,构建了一种新的建树方案,该建树方案从包含所有类的类簇中逐步分离出分离因子最大的类到另外一个类簇,最终形成两个类别数相等的类簇,接着分别递归分解各个类簇,直到各类簇的类别数为1.对照实验表明在训练时间以及识别时间复杂度保持不变的情况下,采用该改进建树方案可有效地提高SVM多分类的识别准确率.When dealing with the algorithm of support vector machine with balanced binary decision tree for multi-class classification, the most difficulty point is to build an efficient balanced binary decision tree . To figure out this key problem , this paper proposes a new scheme to build an efficient balanced binary decision tree which adopts a new scheme of distance measure between classes and in- troduces concepts of divisibility between classes and division factor. The new balanced binary decision tree building scheme separate maximum division factor type from all clusters to another cluster to form two clusters with equal type number, then it iteratively de- compose each cluster individual separately until type number is 1. Finally, the contrast experiment shows that in case of invariance of training time complexity and recognition time complexity , the improved measure can efficiently enhance recognition accuracy.

关 键 词：支持向量机 类间距离度量 类间可分性 分离因子 平衡二叉决策树 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]