Abstract: This paper introduces the concept of pure interval, proposes a new splitting method based on pure intervals reduction to deal with numeric attributes for SPRINT algorithm. The method divides the numeric attributes to many intervals with equal-width histogram, reduces the pure intervals, calculates exactly the minimum gini value in the impure intervals, ensures the accuracy of split result and reduces computation.

Key words: Decision tree, SPRINT algorithm, Pure intervals reduction, Gini index

摘要： 引出了纯区间的概念后，提出了一种基于纯区间归约的数值型属性处理方法对SPRINT算法进行改进。该方法将属性值域用等宽直方图的方法划分为多个区间，对纯区间进行归约，对非纯区间进行精确计算，保证了分裂精度，减小了计算量。

关键词: 决策树, SPRINT算法, 纯区间归约, Gini指数

CLC Number: 

• TP311