一种新型决策树属性选择标准

讨论传统决策树算法中三种常用的基于熵的属性选择标准，提出一种基于属性重要性排序的建立决策树的新方法。该方法在决策树的每个内结点首先依据属性重要性将属性进行排序，然后选择最重要的属性作为分类属性生成决策树，并抽取出规则。与传统的决策树数据分类方法相比，此方法可有效地选择出对于分类最重要的分类属性，增强决策树的抗干扰能力，并提高规则的预测精度。

A New Attribute Selection Criterion of Decision Tree

This paper discusses three common entropy-based attribute selection criteria of the traditional decision tree arithmetic,and presents a new decision tree building method based on attribute importance ranking. The method ranks attributes based on the importance of the attributes in every decision tree interior nodes, and then selects the most important attribute as the ranking attribute to build a decision tree and extracts rules. Compared with the traditional data classification methods used in decision tree, the proposed method can find out the most important attribute efficiently, raises the anti-jamming capacity of decision tree and improves the prediction precision of rules produced.