Abstract: This paper presents a revised decision-tree classification algorithm for customers’ churning, M-AdaBoost cascade of decision tree. By constructing a number of AdaBoosted decision tree classifier, the algorithm produces cascade composition of the decision tree. Experimental results prove the improvement achieved by the new algorithm, and it is better than the result of C4.5 decesion tree, AdaBoost decesion tree and random forest.

Key words: cascade structure, decision tree, false alarming rate

摘要： 提出一种针对客户离网问题的改进决策树分类算法——M-AdaBoost级联决策树。采用级联式的思想构造多个基于AdaBoost决策树分类器，通过设定子分类器的判决信息，组合成级联式决策树。实验结果表明，该方法相对于单一的C4.5决策树、传统的AdaBoost决策树以及随机森林具有更好的分类效果。

关键词: 级联结构, 决策树, 虚警率

CLC Number: 

• TP311.5