Credit Card Fraud Detection Based On Generative Adversarial Nets

Credit card fraud detection is one of the most important means for banks to deal with credit card fraud,and it is also the key work of bank credit card business.In credit card transactions,the amount of normal transaction data is far greater than the amount of abnormal transaction data,so the research on credit card fraud detection is essentially a dichotomy problem of unbalanced data sets,that is,based on the cardholder’s multi-dimensional information,use appropriate big data mining methods to identify and predict whether there is fraud risk.Therefore,improving the performance of machine learning methods in imbalanced data classification problems while ensuring the recognition accuracy of majority and minority classes is of great significance for banking and financial institutions.Based on the credit card transaction data of European credit card holders,this paper balances the data sets using three generation confrontation networks,and constructs a credit card fraud detection model using XGBoost model as the base classifier.Firstly,relevant literature on imbalanced data classification methods was organized,and the feasibility of using generative adversarial networks to process imbalanced data was discussed;Secondly,a feature engineering method based on voting was used to select features from the original data,and 16 features were selected from 30 features for subsequent model construction and training;Then,we use GAN,WGAN and Cramer GAN to balance the credit card transaction data,and use the processed data to train and obtain a credit card fraud detection model based on the generation countermeasure network;Finally,by comparing the performance and classification accuracy of the model built in this paper with the existing model,the feasibility and superiority of the credit card fraud detection model based on the generation countermeasure network are further verified.The research results show that:(1)Among the three credit card fraud detection models based on the generation countermeasure network,the Camera GAN based model has the highest classification accuracy under the condition that the model performance is guaranteed.The F1 value,G-means value,AUC value and AUPRC value are 0.834,0.957,0.968 and 0.803,respectively.(2)Compared with the credit card fraud detection model based on oversampling,the F1 value,G-means value and AUPRC value of the credit card fraud detection model based on the generation of countermeasures network increased by 13.5%,13.7% and 2.33% respectively,which has better model performance and higher classification accuracy.(3)Compared with the models published in SCI and IEEE journals in the past two years,the F1 value and G-means value of the credit card fraud detection model based on Cramer GAN increased by 0.83% and 2.13% respectively,with a higher classification accuracy.Therefore,combined with theoretical analysis and empirical results,it can be shown that: in the credit card fraud detection problem,the credit card fraud detection model proposed in this paper based on the generation countermeasure network has better performance and higher classification accuracy.