人工数据合成法下的通讯客户预警模型
2021-08-06余婉露
余婉露



摘 要: 如何帮助企业提前识别高风险流失客户,已成为许多管理者关心的问题。许多数据挖掘方法用于通讯客户流失案例中时,存在因變量的分布不均匀导致算法精度下降的问题。文章采用人工数据合成法来解决该问题,提出四种客户流失预警模型:GLM-logistic回归模型,GAM-logistic回归模型,Sem-parameter GAM-logistic回归模型和随机森林模型。以AUC和覆盖率-捕获率作为评价指标进行比较,构建出最合适该案例的Sem-parameter GAM-logistic预警模型,以帮助企业减少不必要的客户流失及由此带来的企业损失。
关键词: 人工数据合成法; 预警模型; Sem-parameter GAM-logistic; 覆盖率-捕获率
中图分类号:O213 文献标识码:A 文章编号:1006-8228(2021)07-06-04
Communication customer churn prediction model with synthetic data generation
Yu Wanlu
(Jinshan College of Fujian Agriculture and Forestry University, Fuzhou, Fujian 350002, China)
Abstract: How to help enterprises identify high-risk customer churn in advance has become one of the concerns of many enterprise managers. When many data mining methods are used in communication customer churn cases, the uneven distribution of dependent variables leads to the decline of algorithm's accuracy. In this paper, synthetic data generation is used to solve this problem, and four customer churn early warning models are put forward, i.e. GLM-logistic regression model, GAM-logistic regression model, Sem-parameter GAM-logistic regression model and random forest model. And AUC and coverage rate-capture rate are used as evaluation indexes to build the most suitable Sem-parameter GAM-logistic early warning model for the case, so as to help the enterprise reduce unnecessary customer churn and the losses caused thereby.
Key words: synthetic data generation; prediction model; Sem-parameter GAM-logistic; coverage rate-capture rate
0 引言
随着大数据处理和分析技术的不断发展,客户选择产品以及服务的形式越来越多样化,所以,企业如何对客户数据进行深度挖掘,减少现有客户群流失且发现新的客户群体,显得十分重要。以通讯运营商企业为例,通讯企业想要在日益激烈的市场环境下稳定快速发展,并收获最大经济、社会效益,就离不开高质量的企业客户维系管理[1],因此,通讯客户流失量预测与分析成为各大运营商关注的焦点问题。
数据挖掘技术不断进步,越来越多的客户流失预警模型都用到了数据挖掘技术。在众多预警模型中,常用的数据挖据算法有逻辑回归模型、广义可加模型、支持向量机、决策树、神经网络、随机森林等[2-3]。正确选择以及处理预警模型对模型预测的准确性及效率有着很大影响。
本文深入分析和研究了一些常用客户流失预警模型的相关算法[4],比较各种模型的优势和不足。……
