基于深度神经网络权重集成的客户流失预测  

Customer Churn Prediction Based on Deep Neural Network Weight Ensemble

作　　者：利向晴 夏国恩 张显全[1] 唐琪[1] 叶帅 LI Xiang-qing;XIA Guo-en;ZHANG Xian-quan;TANG Qi;YE Shuai(School of Computer Science and Information Engineering,Guangxi Normal University,Guilin 541004,China;School of Business Administration,Guangxi University of Finance and Economics,Nanning 530003,China)

机构地区：[1]广西师范大学计算机科学与信息工程学院,广西桂林541004 [2]广西财经学院工商管理学院,广西南宁530003

出　　处：《计算机技术与发展》2021年第10期18-23,共6页Computer Technology and Development

基　　金：国家自然科学基金资助项目(71862003);2019年度广西高校中青年教师科研基础能力提升项目(2019KY0661)。

摘　　要：如今许多的企业都面临着客户流失的实质问题,客户流失预测对企业的发展也是尤为重要的。对于客户流失预测的问题也有很多解决方案,传统的机器学习方法存在特征工程对模型效果影响较大的缺点,而深度学习则使得算法不会太依赖于领域专业知识和人工特征提取。为提升预测效果,提出一种深度神经网络权重集成方法来对电信客户流失进行预测,其主要思想是在做深度神经网络(DNN)训练的时候,通过随机加权平均(stochastic weight average,SWA)结合同一网络结构下的不同训练阶段的权重获取集成模型,然后对客户流失进行预测。实验结果表明,相比于深度神经网络训练,结合随机加权平均权重集成的深度神经网络训练时间缩短了2.96倍,同时,准确率、精准率、召回率和F1值均有所提升。Now many enterprises are faced with the problem of customer churn,and customer churn prediction is particularly important for the development of enterprises.There are many solutions for customer churn prediction.Traditional machine learning methods have the disadvantage that feature engineering has great influence on model effect,and deep learning makes the algorithm less dependent on domain expertise and artificial feature extraction.In order to improve the prediction effect,a deep neural network combined with stochastic weights average is proposed to predict customer churn.The main idea is to obtain the integration model by combining the stochastic weight average(SWA)with the weights of different training stages under the same network structure when doing DNN training,and then predict customer churn.The experiment shows that compared with the deep neural network training,the training time of the deep neural network combined with the stochastic weighted average weight integration is shortened by 2.96 times,and the accuracy,precision,recall and F1 are improved.

关 键 词：客户流失 深度学习 深度神经网络 随机加权平均 权重集成 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]