混合模型用于电信客户流失预测解析.doc

混合模型用于电信客户流失预测 Telecom Customer Churn Prediction by Hybrid Model 冉建荣 RAN Jian-Jong; 邵培基 SHAO Pei-Ji; 梁丽琴 LIANG Li-Qin （电子科技大学，成都 610054） (University of Electronic Science and Technology of China, Cheng Du 610054) 摘 要：针对国内省级电信公司客户流失管理实践需要，文章以构建更高效的客户流失预测模型为目标，选用部分数据混合模型方法，以C5.0、Neural Net和Logistic三种算法作为构建混合模型的基础算法。在实证研究过程中，分别采用品牌、地区、网龄和账单作为客户细分标准，构建了不同的客户流失预测混合模型，用命中率和ROC曲线对预测结果比较评估，得到以地区为客户细分标准的混合模型预测效果最佳，以账单层次为客户细分标准的混合模型预测效果其次，以网龄和品牌为客户细分标准的混合模型预测效果较差的结论。因此，作者建议省级电信公司在构建流失预测系统过程中，以客户账单（或客户消费层次）作为细分标准，对各地区客户单独构建流失预测模型；同时，要加强品牌管理，提高各品牌对客户的区隔效果。 Abstract: According to practical needs about customer churn management in provincial telecommunication enterprise of China, this article selects segmented data hybrid model approach and three base arithmetic(C5.0、Neural Net and Logistic) in order to constructing better customer churn prediction model. In course of empirical study, this article constructs four customer churn prediction hybrid models with brand、district、duration and account bill as segmentation variable respectively, and adopts hit ratio and ROC curve to assess prediction effects of the empirical study results. The results show that prediction model of using district as segmentation variable is best; and prediction model of using account bill as segmentation variable is secondary; and prediction model of using duration and brand as segmentation variable is worst. So the author proposes that provincial telecommunication enterprise constructs customer churn prediction model of every district by using account bill as segmentation variable; and strengthens brand management to promote the differentiated effect of every brand. 关键词：客户流失 混合模型 接受者操作特性曲线树 神经网络 逻辑回归 Key words: Customer Churn; Hybrid Model; Receiver Operating Characteristic Curve; Decision Tree; Neural Net; Logistic 0 引言 客户流失是指客户终止与企业的服务合同或转向其它公司提供的服务[1]。据中国工业与信息化部最新统计数据表明，截止2008年11月，中国移动电话用户数达6.34亿户，比去年同期增长19.2%；固定电话用户数达3.48亿户，比去年同期减少5.6%[2]，表明国内市场逐渐趋于饱和。对电信运营企业来说，在这样一个相对成熟的市场中竞争，防御性市场策略变得越来越重要。防御性市