作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2006, Vol. 32 ›› Issue (18): 25-27. doi: 10.3969/j.issn.1000-3428.2006.18.010

• 博士论文 • 上一篇    下一篇

改进RPE算法的神经网络在客户欺诈预测中的应用

王华秋1,2,曹长修2,何 波1,刘祥明2   

  1. (1. 重庆工学院计算机学院,重庆400050;2. 重庆大学自动化学院,重庆400044)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2006-09-20 发布日期:2006-09-20

Application of Neural Networks Based on Modified RPE Algorithm in Customer Fraud Prediction

WANG Huaqiu1,2, CAO Changxiu2, HE Bo1, LIU Xiangming2   

  1. (1. School of Computer Science, Chongqing College of Industrial Technology, Chongqing 400050;2. Automation College, Chongqing University, Chongqing 400044)
  • Received:1900-01-01 Revised:1900-01-01 Online:2006-09-20 Published:2006-09-20

摘要: 客户欺诈在一定程度上抑制了消费,这会妨碍电信运营商和电信用户的亲密度,从而削弱电信运营商的市场竞争力。客户欺诈现象存在非常复杂的多元非线性关系,从统计学角度出发,难以建立预测模型,针对这些问题,提出了基于递推预报误差(RPE)算法神经网络的方法建模,并采用改进的动态遗忘因子方法保证了平稳收敛。实验结果表明,用该算法预测客户欺诈的危险度效果优于BP神经网络模型,具有实用性和有效性。

关键词: 递推预报误差算法, 改进动态遗忘因子, 客户欺诈, 预测模型

Abstract: Customer fraud restrains the consumptions of customers to a certain extent, which might impede good-fellowship between telecom service providers and customers. By that means, this behavior impairs telecom service providers’ market competition. The paper examines customer fraud is a very sophisticate phenomenon which is diverse and nonlinear. From the point of view of statistics, it is difficult to build up prediction model. Aiming at these problems, the paper puts forward a neural network based on recursive prediction error (RPE) algorithm to build up such a model, and adopts modified dynamic oblivious coefficient to ensure stable convergence. Finally the test results show that the proposed neural networks model based on modified RPE algorithm outperforms BP neural networks model. This proves the implementation of the proposed prediction model to be practicable and effective in its application.

Key words: Recursive prediction error algorithm, Modified dynamic oblivious coefficient, Customer fraud, Prediction model