基于改进SMOTE算法和深度学习集成框架的信用卡欺诈检测
Credit Card Fraud Detection Based on Improved SMOTE Algorithm and Deep Learning Ensemble Framework
顾明 1李飞凤 1王晓勇 2郑冬花3
作者信息
- 1. 淮南联合大学经济管理学院,安徽淮南 232038
- 2. 淮南联合大学信息工程学院,安徽淮南 232038
- 3. 广州商学院 信息技术与工程学院,广东 广州 511363
- 折叠
摘要
当前机器学习(ML)算法已经被广泛用于信用卡欺诈检测.然而持卡人线上购物的动态性,以及正常和欺诈交易数据严重不平衡问题,影响了分类器的检测精度.为此,提出了基于深度学习集成框架的信用卡欺诈检测方法.首先,通过改进的合成少数类过采样(SMOTE)算法,解决信用卡数据集中欺诈交易和正常交易数量严重不平衡问题.其次,构建堆栈式深度学习集成框架,使用双向长短时记忆网络(Bi-LSTM)和门控循环单元(GRU)作为基础分类器,并通过多层感知机(MLP)作为元分类器,结合集成学习和深度学习的优点提高信用卡欺诈检测率.在公开数据集上的实验结果表明,所提深度学习集成方法与改进SMOTE算法相结合,分别实现了99.57%和99.82%的灵敏度和特异性结果,优于其他先进的信用卡欺诈检测算法.
Abstract
Machine learning(ML)algorithms have been widely employed in credit card fraud detection.However,the dynamic nature of online shopping by cardholders,coupled with the severe imbalance between normal and fraudulent transaction data,adversely affects the detection accuracy of classifiers.In response,this study proposes a credit card fraud detection method based on a deep learning ensemble framework.Firstly,an improved Synthetic Minority Over-sampling Technique(SMOTE)algorithm is introduced to address the significant imbalance between the quantities of fraudulent and normal transactions in credit card datasets.Secondly,a stacked deep learning ensemble framework is constructed,utilizing Bi-directional Long Short-Term Memory(Bi-LSTM)and Gated Recurrent Unit(GRU)net-works as base classifiers,and employing a Multi-Layer Perceptron(MLP)as a meta-classifier.This combination harnesses the advantages of both ensemble learning and deep learning to enhance the effectiveness of credit card fraud detection.The experimental results on the public dataset indicate that the proposed deep learning ensemble method,combined with the improved SMOTE algorithm,achieved sensitivity and specificity results of 99.57%and99.82%,re-spectively.This outperforms other state-of-the-art algorithms for credit card fraud detection.
关键词
信用卡欺诈检测/机器学习/深度学习/合成少数类过采样/双向长短时记忆网络/门控循环单元Key words
Credit card fraud detection/Machine learning/Deep learning/Synthetic Minority Over-sampling Tech-nique/Bi-directional Long Short-Term Memory/Gated Recurrent Unit引用本文复制引用
基金项目
2023年度安徽省科研编制计划科学研究重点项目(2023AH051161)
出版年
2024
NETL
NSTL
万方数据