竞争选择分裂属性的决策树分类模型

构建决策树分类器关键是选择分裂属性。通过分析信息增益和增益比率、Gini索引、基于Goodman-Kruskal关联索引这三种选择分裂属性的标准,提出了一种改进经典决策树分类器C4.5算法的方法(竞争选择分裂属性的决策树分类模型),它综合三种选择分裂属性的标准,通过竞争机制选择最佳分裂属性。实验结果表明它在大多数情况下,使得不牺牲分类精确度而获得更小的决策树成为了可能。

A Decision-Tree Classifier Model of Competition in Choosing Split Attribute

The construction of decision-tree is centered on the selection algorithm of an attribute that generates a partition of the subsets of the training database that is located in the node about to be split.On the basis of analyzing three techniques for choosing the splitting attributes including the entropy gain and the gain ratio,the gini index and Goodman-Kruskal association index,propose a strategy to improve on classical decision-tree classifier C4.5 arithmetic(a decision-tree classifier model of competition in choosing split attribute).Experimental results show it is possible,in most cases,to obtain smaller decision trees without sacrificing accuracy.