摘要
决策树归纳的两个重要阶段是数据表示空间的简化和决策树的生成。在将训练集的不一致率控制在某一阈值的前提下,减少实例的属性个数和各个属性的取值个数保证了决策树方法的可行性和有效性。本文在Chi2算法的基础上运用它的一种变形进行属性取值离散化和属性筛选,然后运用算术运算符合并取值个数为2或3的相邻属性。在此基础上生成的决策树具有良好的准确性。实验数据采用的是一个保险公司捐献的数据集。
The simplification of training dataset representation and the generation of decision trees are two critical phases in decision tree induction. On the condition of bringing the inconsistency rate under a threshold, reducing the attribute number and the different value number of each attribute assures the feasibility and effectiveness of the decision tree learning method. In this paper, a variation of the Chi2 algorithm is proposed to perform attribute discretization and selection. The decision tree generated in the further steps offers a good classification accuracy. Our experiment is based on a data set donated by an insurance company from the real world.
出处
《计算机工程与科学》
CSCD
2007年第10期47-49,共3页
Computer Engineering & Science
基金
天津市自然科学基金资助项目(033610811)
天津市"十五"教育科学规划重点课题(YSO17)
关键词
决策树
Chi2的变形
离散化
筛选
decision tree
variation of Chi2
discretization
selection
