东北大学学报:自然科学版 ›› 2019, Vol. 40 ›› Issue (2): 180-185.DOI: 10.12068/j.issn.1005-3026.2019.02.006

• 信息与控制 • 上一篇    下一篇

基于改进GA-BP的移动通信用户流失预测算法

于瑞云, 薛林, 安轩邈, 夏兴有   

  1. (东北大学 软件学院, 辽宁 沈阳110169)
  • 收稿日期:2017-11-15 修回日期:2017-11-15 出版日期:2019-02-15 发布日期:2019-02-12
  • 通讯作者: 于瑞云
  • 作者简介:于瑞云(1974-),男,辽宁丹东人,东北大学教授.
  • 基金资助:
    国家自然科学基金资助项目(61672148,61502092); 教育部-中国移动科研基金资助项目(MCM20160201); 辽宁省百千万人才工程项目(201514).

Mobile Communications Customer Churn Prediction Algorithm Based on Improved GA-BP Network

YU Rui-yun, XUE Lin, AN Xuan-miao, XIA Xing-you   

  1. School of Software, Northeastern University, Shenyang 110169, China.
  • Received:2017-11-15 Revised:2017-11-15 Online:2019-02-15 Published:2019-02-12
  • Contact: XUE Lin
  • About author:-
  • Supported by:
    -

摘要: BP神经网络(BPNN)模型对移动通信用户流失的预测有较好的效果,但其全局搜索能力相对较弱,对初始网络权重非常敏感,因此本文通过对用户通信行为的分析,提出一种基于改进GA-BP的移动用户流失预测算法:用改进的遗传算法对BPNN的权值和阈值进行初始化,从而提高预测模型的准确率.改进的遗传算法采用一种自适应的交叉概率和变异概率计算策略,提高了遗传算法寻找全局最优解的能力.通过对比实验发现,本文构建的移动用户流失预测模型,在预测准确率上有着很好的表现.

关键词: 移动通信, 行为分析, 用户流失, BP神经网络, 遗传算法

Abstract: A customer churn prediction model based on BP neural network(BPNN)has achieved well enough results. However, it has relatively weak global search ability and is very sensitive to the initial network weights. A prediction algorithm based on improved genetic algorithm(IGA)and BPNN(IGA-BP)is proposed by analyzing users’communication behavior, where the weights and thresholds of BPNN are initialized with IGA, thus improving the accuracy of the prediction model. The improved algorithm adopts a self-adapting probability of crossover and mutation, which enhances the global optimum search ability of GA. The proposed IGA-BP model has obvious improvement on customer churn prediction, compared with existing algorithms.

Key words: mobile communication, behavior analysis, customer churn, BP neural network, genetic algorithm

中图分类号: