Abstract: This article analyzes the ID3 decision tree algorithm which is widely used in classifying and forecasting，and points out some defections of this algorithm such as value deviation and operation inefficient and so on.On base of the analysis of ID3，an improved ID3 algorithm is presented to apply it to forecaste the customer’s churn in a communication company.The improved algorithm overcomes the value deviation through the attribute weighting，obviously improves the operation efficiency using the entropy function’s recursion and by looking up the table of dual entropy functions.The application result indicates that the improved algorithm obviously enhances the performance.

Key words: decision tree, customer’s churn, entropy function

摘要： 对分类预测中广泛使用的ID3决策树算法进行了分析，指出了该算法的取值偏向性以及运算效率不高等缺点，在此基础上提出了一种改进的ID3算法并将其应用于某移动通信公司的客户流失预测。改进的算法通过属性加权克服取值偏向性，运用熵函数的递推性质并通过二元熵函数查表法显著地提高运算效率。应用结果表明，提出的改进算法性能明显改善。

关键词: 决策树, 客户流失, 熵函数