信用卡欺诈风险识别综合实验设计  

Comprehensive experimental design of credit card fraud risk identification

作　　者：谢天保[1] 齐德伟 陈梦圆 贾臻 XIE Tianbao;QI Dewei;CHEN Mengyuan;JIA Zhen(School of Economics and Management,Xi’an University of Technology,Xi’an 710054,China)

机构地区：[1]西安理工大学经济与管理学院,陕西西安710054

出　　处：《实验技术与管理》2022年第3期200-204,共5页Experimental Technology and Management

基　　金：西安理工大学教育教学改革研究项目(XJY2007)。

摘　　要：该文针对信用卡欺诈客户数据集极不平衡的特点,设计了SMOTE、Borderline SMOTE、ADASYN、SMOTENC四种采样算法,对数据集进行均衡处理,并结合逻辑回归、支持向量机、随机森林、多层神经网络构建欺诈风险识别模型,最后通过准确率、AUC、精确率、召回率和F1等指标对识别模型效果进行评判。实验结果表明,SMOTENC采样方法与随机森林模型相结合构建的识别模型效果最好,准确率达到99%,可为银行进行客户欺诈风险判别提供支撑。Aiming at the extremely unbalanced data set of credit card fraud customers,this paper designs four sampling algorithms:SMOTE,Borderline SMOTE,ADASYN,SMOTENC to balance the data set,and constructs a fraud risk recognition model which combines logistic regression,support vector machine,random forest and multilayer neural network.Finally,the effect of this recognition model is evaluated by indexes such as accuracy,AUC,accuracy recall rate and F1 score.The experimental results show that the recognition model constructed by the combination of smotenc sampling method and random forest model has the best effect,with an accuracy of 99%,which can provide support for banks to distinguish customer fraud risk.

关 键 词：非均衡数据集 SMOTE 欺诈客户 欺诈风险识别 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论] G642.0[自动化与计算机技术—计算机科学与技术]