基于本质特征和网络特征的信用卡欺诈检测  

Fraud Detection of Credit Card Based on Essential Characteristics and Network Characteristics

作　　者：张燕[1] 

机构地区：[1]宁夏工商职业技术学院,银川750021

出　　处：《微型电脑应用》2016年第12期72-77,共6页Microcomputer Applications

摘　　要：由于信用卡欺诈检测是一种不合规则的预测任务,需要专门方法来处理并预测,提出一种基于本质特征和网络特征的检测方法,以满足自动化和实时处理的要求。提出的方法结合了两种重要特征,即利用新近度—频率—货币值(RFM)的基本原理,由外来交易和顾客消费历史派生出本质特征;采用信用卡持有人和商家的网络为每个网络对象派生出依赖猜测分数的网络特征。然后将这些特征提供给成熟的学习方法。本文评估了逻辑回归、神经网络和随机森林模型。结果表明本质特征和网络特征的结合产生了最佳执行结果,获得的ROC曲线下面积(AUC)高于0.98。且提出的方法还能够精确地从一系列欺诈交易中挑选出第一笔交易。As fraud detection of credit cards is a kind of irregular prediction task and it needs special methods to process and predict, a detection method based on network characteristics and essential characteristics is proposed to meet the requirements of automation and real-time processing. Two important characteristics are combined in the proposed method, which uses the basic principle of Re- cency-Frequency-Monetary（RFM）, and then the essential characteristics is derived by foreign trade and consumer consumption his- tory. The other is network characteristics, in which the guessing score for each network object is derived by the network of credit card holders and merchants. Then these features are provided to mature learning methods. Logistic regression, neural network and random forest model are evaluated in this paper. The results show that the combination of essential features and network characteristics has produced the best results, and the AUC score is higher than 0.98. And the proposed method is also able to pick out the first transac- tion from a series of fraudulent transactions accurately.

关 键 词：信用卡欺诈检测 预测 本质特征 网络特征 新近度-频率-货币值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]