基于混合采样和强化学习的信用卡欺诈检测模型  

Credit card Fraud Detection Model Based on Mixed Sampling and Reinforcement Learning

作　　者：郑越 代琪 施永辉 韩阳 陈丽芳 ZHENG Yue;DAI Qi;SHI Yong-hui;HAN Yang;CHEN Li-fang(College of Science,North China University of Science and Technology,Tangshan Hebei 063210,China;Discipline Construction Department of North China University of Technology,Tangshan Hebei 063210,China;Key Laboratory of Data Science and Application of Hebei Provincial,Tangshan Hebei 063210,China)

机构地区：[1]华北理工大学理学院,河北唐山063210 [2]华北理工大学学科建设处,河北唐山063210 [3]河北省数据科学与应用重点实验室,河北唐山063210

出　　处：《华北理工大学学报（自然科学版）》2024年第3期131-140,共10页Journal of North China University of Science and Technology：Natural Science Edition

基　　金：国家自然科学基金面上项目(52074126):基于高炉冶炼过程大数据深度挖掘的炉温智能管控模型研究。

摘　　要：针对信用卡欺诈检测中存在数据不平衡以及模型超参数不能自动调优,导致检测精度低的问题,综合考虑数据平衡处理和参数优化两个关键因素,构建了一种基于混合采样和强化学习的信用卡欺诈检测模型。所提模型采用Kmeans聚类算法对多数类数据以及少数类数据分别聚类,多数类数据保留具有代表性的数据,少数类数据采用生成对抗网络生成数据,从而达到数据集平衡。同时,提出了一种改进的Q-learning算法,通过优化算法中奖励值的计算和Q-table的更新方式,以加快算法的收敛速度,然后采用改进的Q-learning算法优化XGBoost模型中的参数,以提高模型的检测效果。实验结果表明:相比于传统的欺诈检测模型,该模型具有更优的检测效果,更适合应用于信用卡欺诈检测领域。A credit card fraud detection model based on mixed sampling and reinforcement learning was proposed to address the issues of data imbalance and inability to automatically tune model hyperparameters,which lead to low detection accuracy.This model adopts the K-means clustering algorithm to cluster majority and minority class data separately.Representative data from the majority class were retained while minority class data were generated using generative adversarial networks to achieve dataset balance.Additionally,an improved Q-learning algorithm was proposed,which accelerates the convergence speed of the algorithm by optimizing the calculation of rewards and the updating mechanism of the Q-table.Then,the improved Q-learning algorithm was used to optimize the parameters of the XGBoost model to enhance the detection performance.Experimental results demonstrate that compared to traditional fraud detection models,this model exhibits superior detection performance and is more suitable for application in the field of credit card fraud detection.

关 键 词：强化学习 生成对抗网络 混合采样 欺诈检测 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]