决策树中数值型属性分裂的研究  

Research on Numeric Attributes Splitting in Decision Trees

作　　者：刘友军[1] 汪林林[1] 

机构地区：[1]重庆邮电学院,重庆400065

出　　处：《计算机科学》2006年第5期241-243,共3页Computer Science

摘　　要：在介绍了现有数值型属性分裂方法的基础上,引出了纯区间的概念,提出了一种基于纯区间归约的数值型属性分裂方法。该方法将属性值域用等宽直方图的方法划分为多个区间,对纯区间和非纯区间分别处理。理论分析和实验结果表明该方法在保证了分裂精度的同时,减小了搜索空间。This paper introduces methods about numeric attributes splitting and the concept of pure interval, propose a new splitting method based on pure intervals reduction. The method divides the numeric attributes to many intervals with equal-width histogram, uses different methods to deal with the pure and impure intervals. Theoretical analysis and experimental results showe that the proposed method ensures the accuracy and narrowes the search space.

关 键 词：决策树 数值型属性 纯区间归约 Gini指数 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]