电信业移动客户流失预测C-N模型研究

【作者】 周晓华；

【导师】 邵培基；

【作者基本信息】 电子科技大学 ， 管理科学与工程， 2007， 硕士

【摘要】 近年来,移动电话用户的数量呈现出了稳步增长的局面。客户已成为各大电信运营商宝贵的资源,电信运营商之间的竞争也日趋激烈。如何有效地避免客户流失,保留客户,制定科学合理的营销和服务策略从而最大程度地降低客户的流失率,已成为目前运营商需要解决的一个问题。电信运营商经过多年的发展,已经拥有许多成熟的数据库系统,积累了大量业务数据。然而面对海量的数据,要想从中获取有用信息,预测出客户在未来一段时间内的流失倾向,以及如何更有针对性地对即将流失的客户采取措施进行挽留,传统的基于客户行为的分析方法已经不能满足这样的需求。国外电信业已有将数据挖掘应用于客户流失预测和流失控制的成功案例。国内电信企业目前主要集中在通过建立客户关系管理系统来进行客户维系挽留,根据现有文献,尚未有效地对客户流失进行预测。对客户流失预测的研究主要采用单个数据挖掘算法建立预测模型。这些方法往往只给出客户流失概率,缺乏对实际应用的指导。因此,本文在前人研究的基础上,结合目前电信企业客户维系管理现状,构建了基于决策树和神经网络的电信客户流失预测C-N混合模型,对电信客户流失进行预测。本文采用理论研究和实证研究相结合的方法,以中国联通G分公司的客户数据为基础,构建了基于决策树和神经网络的C-N混合模型。文中详细阐述了包括属性的选择、数据的准备、模型的构建以及模型的评价和应用的整个过程。本文采用数值化指标和图形化指标对模型进行评价,使得评价更加客观,研究结果表明混合模型在预测精度和命中率方面都有所改善,从整体性能来看也要优于电信业现有的客户流失分析方法。最后,本文利用C-N模型的结果,在分析中国联通G分公司在网用户未来一个月的流失概率的基础上,探讨了流失概率高的客户月消费额、在网时长、性别、网内网外通话次数等属性,总结了客户流失的原因,并提出了相应的客户挽留措施。本文提出的基于决策树和神经网络的移动客户流失预测C-N混合模型可以为我国电信行业客户流失分析提供一定的参考。更多还原

【Abstract】 Mobile phone users have been rising in a stable way in recent years. Customers have become valuable resources of telecommunications operators, competition between the telecommunications operators become drastic. How to avoid customers churn, retain customers, make scientific and rational marketing and services strategy to reduce customer churn ratio at the greatest extent, has become one of the focus telecommunication operators should pay attention to.After years of development, telecommunication operators have many mature database systems, which accumulated large amounts of business data. However, faced with massive data, if we want to obtain useful information, to predict churn tendency of customer in the near future, and how to take proper measures to retain customers, traditional analysis method which Based on the customer behavior has been unable to meet this demand.Some Foreign telecom companies have already used data mining for customer churn predicting. Domestic telecommunication enterprises focus on establishing customer relationship management system to maintain and retain customers, few predict customer churn according to existing literature.The research on customer churn prediction focus on using single data mining method to build predictive model, these methods always only give the probability of customer churn, and they don not represent better effect in practice. Based on former research, combining the mobile communications industry status in customer hold management, this paper builds C-N hybrid model for customer churn prediction using decision tree and neural network method .In this paper, a theoretical research and empirical study method is used. Based on China Unicom G branch’s customer data, this paper builds C-N hybrid model, including a detailed explanation of the whole process such as attributes choosing, data preparation, construction of the model and model evaluation and application. In this paper, a more reasonable evaluation method - numerical indicators and the graphic indicators are used to evaluate the result of the model. The result indicates that the hybird model has better accuracy and hit rates. Meanwhile, the C-N model presents better results than the existing method used by telecom industry at home.Finally, using the results of the C-N model, this paper analyses the probability of the G Unicom’s customer churn in the next month, and the churning customer’s characteristics, such as average fee, length of service, gender, total number of outgoing calls and so on, and sum up the reason for the loss of customers, and give the corresponding measures to retain customers. The customers predicting hybird mobile proposed in this paper may provide some references to the analysis of the customer churn in the telecommunication industry.更多还原