组合模型在电信客户流失预测中的应用  

Application of Combinatorial Model in Telecom Customer Churn Prediction

作　　者：沈江明 孙凯 曾志勇[3,4] SHEN Jiang-ming;SUN Kai;ZENG Zhi-yong(Yunnan Corporation,China Telecom Group,Kunming Yunnan 650000,China;School of Statistics and Mathematics,Yunnan University of Finance and Economics,Kunming Yunnan 650000,China;School of Information,Yunnan University of Finance and Economics,Kunming Yunnan 650000,China;Yunnan Universities Research Center for Data Operation and Management Engineering,Kunming Yunnan 650000,China)

机构地区：[1]中国电信股份有限公司云南分公司,云南昆明650000 [2]云南财经大学统计与数学学院,云南昆明650000 [3]云南财经大学信息学院,云南昆明650000 [4]云南省高校数据化运营管理工程研究中心,云南昆明650000

出　　处：《通信技术》2020年第6期1575-1580,共6页Communications Technology

摘　　要：针对电信客户流失模型的构建,提出了基于不均衡数据处理与组合模型相结合的集成方法。按固定比例同时对数据集中多数类样本和少数类样本抽样,形成一个新的子数据集,重复该过程并训练多个基分类器;将基分类器进行线性组合,利用Lagrange函数求解组合模型的系数。利用某企业宽带客户行为数据训练模型,进行隔月预测。实验结果表明:该方法相对于各单模型,在F1值和对少数类的预测命中率上分别提升了2.3%和2.1%,可以帮助企业制定挽留方案。Aiming at the construction of telecom customer churn model,an integration method based on the combination of unbalanced data processing and composite model is proposed.According to the fixed proportion,the majority samples and the minority samples in the data set are sampled at the same time to form a new sub-data set.The process is repeated,multiple base classifiers are trained,the base classifiers are combined linearly,and the Lagrange function is used to solve the coefficients of the combined model.Then,certain enterprise’s broadband customer behavior data training model is used to make predictions every two months,and the experimental results indicate that this method has improved the F1 value and the predicted hit rate for minority classes by 2.3%and 2.1%,respectively,which can help companies formulate their retention plans.

关 键 词：电信客户流失 LAGRANGE函数 隔月预测 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]