基于贝叶斯网络的电信客户流失预测分析  

A Data Mining Method of Forecasting the Customer Churn in Telecom Company Based on Bayesian Network

作　　者：叶进[1] 程泽凯[2] 林士敏[2] 

机构地区：[1]桂林电子工业学院通信与信息工程系,桂林541004 [2]广西师范大学计算机科学系,桂林541004

出　　处：《计算机工程与应用》2005年第14期212-214,共3页Computer Engineering and Applications

基　　金：清华大学智能技术与系统国家重点实验室开放课题资助(编号:99002)

摘　　要：电信客户流失分析常用的数据挖掘方法有自动聚类、决策树和人工神经网络,它们是采用数据本身来训练模型的,没有利用先验知识。电信客户流失是由客户心理、服务质量和对手竞争等诸多复杂的因素造成的,利用这些已有的先验知识,可以提高预测的精度。该文根据先验知识选取分析变量,采集样本数据,通过贝叶斯网络的结构学习和参数学习,建立客户流失模型并进行客户流失趋势预测,取得了比标准数据集更准确的结果,该结果和决策树方法的预测结果相比还具有较大的优势,说明贝叶斯网络是分析客户流失等不确定性问题的有效工具。Generally used methods of analysis churn of telecom company are Automatic cluster detection,Decision Tree and Artificial Neural Network,which are based on the data itself.However the customer churn is led by many complex reasons such as customer mentality,service quality and competition.If we combine domain knowledge,the accuracy of forecast could be improved.By our experiment some variables are selected as network node and part of data is selected as sample,this paper discusses how to build the model of customer churn and forecast the churn rate with the structure learning and parameters learning of Bayesian network.The result of experiment is more accurate than other experiment which standard data is selected.At the mean time,the advantage of the result have been shown in the comparison with the result of decision tree.It is proved that Bayesian network is a good method to solve uncertain problem such as analysis of customer churn.

关 键 词：数据挖掘 机器学习 贝叶斯网络 贝叶斯方法 客户流失预测 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]