客户流失预测理论与实证研究

项目名称： 客户流失预测理论与实证研究

项目编号： No.70801021

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 数理科学和化学

项目作者： 夏国恩

作者单位： 广西财经学院

项目金额： 16.8万元

中文摘要： 传统客户流失预测研究没有统一的模型结构，并且所采用的方法在理论上不能同时保证预测模型的泛化能力和可解释性。本项目从理论研究出发，合理的拓展了传统客户流失预测模式，由此从理论上重新审视预测模型结构，并分别考察引入属性选择、特征提取和特征选择方法、可解释性的基于结构风险最小化方法、集成学习方法后，预测模型的精度和解释性的变化。通过利用实际数据对提出的理论成果进行检验，进一步论证了所建立模型的有效性和可靠性。同时，还基于竞争对手反击效应存在性建立了客户挽留价值计算模型；引入隐马尔科夫和动态规划技术，提出了个体客户全生命周期价值动态模型；利用模糊测度加强不确定性特征信息的量化，建立基于模糊测度的k近邻分类算法（FM-KNN）；通过比较客户的平均等待时间，研究了基于最大熵方法的稳定队长分布；研究了制造商与两个内部客户（零售商）供应链协作的基于数量折扣的混合契约问题。上述研究，完善了客户流失预测理论，也为解决企业中客户流失问题提供了一定的理论依据和实践。

中文关键词： 客户流失；预测模型；精度；解释性

英文摘要： many empirical studies show that no uniform model exists in the tradition researches of customer churn prediction, and the theory of the old methods can not synchronously guarantee the good generalized performance and explanation of the prediction model. Based on surveying newly-established prediction model structure from the theoretical research and reviewing the change of model precision and explanation by using attribute selection, feature extraction, feature selection, explanative structure risk method and integration learning, this task logically develops the traditional pattern of customer churn prediction. By testing the proposed theory with the real data, it shows that the building models are valid and dependable. At the same time, the customer detainment value model is built based on the rival counterattack effect; by using the hidden markov model and dynamic programming thechnology, it proposes the dynamic model of individual customer's CLV; it presents the fuzzy measures K-Nearest Neighbor(FM-KNN), which applies fuzzy measures to strengthen the quantitative uncertainty characteristic information; by comparing the expected waiting time, it studies the stationary queue-length distribution based on the maximum entropy; The compound contract of based on quantity-discount schedule is proposed for the supply chain coordination of one manufacturer and two cooperative retailers. The above researches perfect customer churn prediction theory and thus provides certain theoretic support as well as practical value for companies to deal with customer churn problems.

英文关键词： customer churn;prediction model; precision;explanation