基于随机森林CART特征选择改进算法的电信客户流失预测模型  

Telecom customer churn prediction model based on improved random forest cart feature selection algorithm

作　　者：乔健 诸佳慧 严康桓 QIAO Jian;ZHU Jia-hui;YAN Kang-huan(Shanghai Information Network Co.,Ltd.,Shanghai 200081,China;Fudan University,Shanghai 200433,China;Shanghai Branch of China Telecom Co.,Ltd.,Shanghai 200085,China)

机构地区：[1]上海市信息网络有限公司,上海200081 [2]复旦大学,上海200433 [3]中国电信股份有限公司上海分公司,上海200085

出　　处：《电信工程技术与标准化》2022年第3期78-82,共5页Telecom Engineering Technics and Standardization

摘　　要：客户流失预测能够帮助运营商制定有针对性的挽留营销政策,对提高竞争力和营业收入有重要意义。本文针对随机森林算法在数据和类别不平衡情况下预测准确率下降的问题,在随机森林CART分类树算法的特征选择过程中引入客户生命周期价值指标,降低了不平衡情况下的基尼系数和模型的不纯度。对电信业客户基本信息、行为数据和交互数据进行数学挖掘和建模,实验结果表明,新改进算法在不平衡情况下可以对潜在流失客户群的特征进行预测,能有效提升客户流失预测模型的准确率,精确评估高价值客户流失临界点,从而快速计算出挽留成本和收益。Customer loss prediction can help operators formulate targeted retention marketing,which is of great signifi cance to improve competitiveness and operating revenue.In this paper,aiming at the problem that the prediction accuracy of random forest algorithm declines in the case of unbalanced data and categories,the customer life cycle value is introduced into the feature selection of random forest cart classifi cation tree algorithm for improvement,which reduces the Gini coefficient and the impure of the model in the case of unbalanced data.The experimental results show that the new improved algorithm can predict the characteristics of potential customer churn under unbalanced conditions,effectively improve the accuracy of customer churn prediction model,accurately evaluate the critical point of high-value customer churn,and quickly calculate the retention cost and revenue.

关 键 词：随机森林 分类回归树 基尼系数 客户生命周期价值 数据建模 

分 类 号：TN915[电子电信—通信与信息系统]