电信运营企业客户流失预测与评价研究

发布时间：2018-05-21 03:43

本文选题：电信运营企业 + 客户流失　；参考：《哈尔滨工程大学》2013年博士论文

【摘要】：电信运营企业客户流失是一个受多因素影响的复杂问题，尤其是2008年以后我国电信业针对3G牌照的发放又进行了新一轮的电信重组，全业务运营下的三大运营企业从此展开了激烈的客户市场竞争。由于我国移动客户群体庞大，中低端客户在不同运营企业间流动性强，因此，针对客户流失的成因分析和建立客户流失预测模型具有重要的理论价值和现实意义。本文详细分析了国内外学者在客户流失领域的研究成果，探讨了客户流失的影响因素和客户流失预测的方法。通过对3G时代电信运营环境的分析，总结了国内外电信运营企业客户流失的现状，并从电信运营环境角度、运营企业流失客户数据统计分析角度深入研究了电信运营企业客户流失的成因，归纳得到客户流失成因的8种类型。据此，基于数据挖掘和客户价值的理论和方法，研究了BP神经网络算法、支持向量机算法、C5.0决策树算法在客户流失预测上的应用，为了获得更好的预测效果，构建了Lagrange组合预测模型和基于客户价值的预测模型。重点就以下问题进行了研究：在广泛研究和借鉴国内外相关数据挖掘理论及成果的基础上，探讨了电信运营企业的客户构成，，深入分析了客户流失与流失客户的概念、以及客户流失的现象与特征，从而梳理给出三户关系模型。对构建模型的客户属性进行了分类，即原始属性与衍生属性。以往对电信客户流失预测的研究都是采用客户消费行为、个人信息、缴费信息等原始属性数据，这些原始属性数据很难真实地反映客户流失的行为；加入了衍生属性，如：月租标志、呼转标志、账户余额标志、充值行为标志等，其数据集能更好的预测客户流失，使得预测的命中率更高，计算的客户价值更具研究意义。通过分析客户协议数据、消费行为数据和账单数据得出与客户流失密切相关的属性集，根据获取运营企业数据的难易程度，建立了客户流失预测指标体系，并基于数据挖掘算法建立了Lagrange组合预测模型。针对客户流失预测问题的研究，选择了数据挖掘的三种经典算法（BP、SVM、C5.0）构建了单一客户流失预测模型，并通过对模型的评估显示，任意单一模型都没有最优。据此借助Lagrange函数求极值的思想构建了客户流失的组合预测模型，其预测效果比单一模型更理想。提出二维度预防客户流失的方法，即基于Lagrange的客户流失组合预测与基于客户价值的流失客户评价。根据组合预测模型预测得到的客户流失名单是否有挽留的价值，或者说是否有对这样的客户有再投入成本挽留的必要，取决于该客户对运营企业是否是有价值客户，并依据这两种途径的预测结果，再分析客户流失的根本原因。最后，通过对客户流失成因的分析，以及对客户流失预测模型的评估，提出电信运营企业降低客户流失的措施与建议。
[Abstract]:Customer churn is a complex problem affected by many factors, especially after 2008, the telecom industry of our country has carried out a new round of telecom reorganization aiming at the issue of 3G license. All-business operation under the three major operating enterprises from then on launched a fierce customer market competition. Because of the huge mobile customer group and the strong liquidity among the middle and low end customers in different operating enterprises, it is of great theoretical value and practical significance to analyze the causes of customer turnover and to establish a customer churn prediction model. In this paper, the research results of domestic and foreign scholars in the field of customer churn are analyzed in detail, and the influencing factors of customer churn and the method of customer churn prediction are discussed. Based on the analysis of telecom operating environment in 3G era, this paper summarizes the current situation of customer churn in telecom operation enterprises at home and abroad, and from the point of view of telecommunication operation environment, In this paper, the causes of customer churn in telecom operation enterprises are studied from the point of view of statistical analysis of customer churn data, and eight types of customer churn are concluded. Based on the theory and method of data mining and customer value, this paper studies the application of BP neural network algorithm, support vector machine algorithm and C5.0 decision tree algorithm in customer churn prediction. Lagrange combination forecasting model and customer value based forecasting model are constructed. Research focused on the following issues: On the basis of extensive research and reference of relevant data mining theories and achievements at home and abroad, this paper probes into the customer structure of telecom operation enterprises, deeply analyzes the concepts of customer churn and customer churn, as well as the phenomena and characteristics of customer churn. In order to sort out the three-family relationship model. The customer attributes of the building model are classified, that is, the original attributes and the derived attributes. In the past, the research of telecom customer churn prediction used the original attribute data, such as customer consumption behavior, personal information, payment information and so on. These original attribute data can hardly truly reflect the behavior of customer churn. Such as: monthly rent sign, call mark, account balance mark, recharge behavior mark, etc., its data set can better predict customer churn, make forecast hit ratio higher, calculate customer value more research significance. Through the analysis of customer agreement data, consumer behavior data and billing data, the attribute set which is closely related to customer churn is obtained. According to the degree of difficulty in obtaining operation enterprise data, a customer churn prediction index system is established. Based on the data mining algorithm, the combined prediction model of Lagrange is established. In order to solve the problem of customer churn prediction, three classical algorithms of data mining are selected to construct a single customer churn prediction model. The evaluation of the model shows that there is no optimal model for any single model. Based on the idea of calculating extreme value of Lagrange function, the combined forecasting model of customer churn is constructed, and its prediction effect is more ideal than that of single model. This paper presents a two-dimensional method to prevent customer churn, that is, the combination prediction of customer churn based on Lagrange and the evaluation of customer churn based on customer value. Whether the customer churn list based on the combination forecasting model has the value of retention, or whether it is necessary to retain such a customer at a cost, depends on whether the customer is a valuable customer to the operating enterprise. And according to the forecast results of these two ways, the root cause of customer turnover is analyzed again. Finally, through the analysis of the causes of customer churn and the evaluation of customer churn prediction model, the paper puts forward the measures and suggestions to reduce customer churn in telecom operation enterprises.
【学位授予单位】：哈尔滨工程大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：F274;F626

【参考文献】