摘要
决策树是一种采用分治策略的聚类分析方法,构建决策树的关键是选择合适的属性。传统的决策树通常从最大化信息熵的角度来构造,不能对属性的分类能力进行足够好的区分。对传统的决策树生成算法的不足,本文提出了一种基于马氏距离的决策树生成算法。算法使用马氏距离来区分不同特征属性子集的分类能力。试验结果表明,基于度量的决策树的性能优于传统的决策树。
Decision tree is a kind of cluster analysis method using divide- and- conquer strategy, the key of making up of decision tree is to choose appropriate attribute. Commonly we construct conventional decision tree from the angle of maximizing entropy of information which can not preferably differentiate the classifying ability of attribute. Contraposing the deficiency of the creating arithmetic of conventional decision tree, this paper proposes a new creating arithmetic of decision tree based on Mahalanobis metric. By using Mahalanobis metric, it can distinguish dassif-ying ability of different feature subsets. Results indicate that the performance of the metric- based decision tree is some better than that of the conventional decision tree.
出处
《计算机与数字工程》
2006年第4期46-47,共2页
Computer & Digital Engineering
关键词
马氏距离
特征子集
决策树
Mahalanobis metric, Feature subset, Decision tree
