Research On Credit Card Fraud Detection Model Based On KBX Integrated Learning

The consumer philosophy of "borrowing from the future to enjoy the blessings of today" has become the mainstream concept of many young people in today’s society.The introduction of credit cards has brought great convenience to people’s daily lives,but at the same time the phenomenon of illegal fraud in China in the credit card business is increasing,causing significant economic losses to the state,society and individuals.Such operational behaviour has seriously disrupted the normal operation of the financial market,so it is of great importance to establish a credit card financial anti-fraud model to promote the healthy development of the industry.This paper presents a detailed summary of the problems and sample imbalance in credit card fraud detection,using credit card transaction data collected by financial institutions as a case study.In this paper,we propose(1)a credit card fraud detection model based on MHA-DNN-Focal.This model combines the idea of multi-headed attention mechanism with Focal Loss function,which not only solves the problem of sample imbalance,but also avoids the problem of discarding potentially valuable features in processing the dataset.(2)A credit card fraud detection model based on KBX integrated learning.This scheme first uses the technique of under-sampling(Under Sampling)and synthetic few-sample oversampling combined with under-sampling(SMOTE+Under Sampling)for imbalance processing,and then compares the advantages and disadvantages of the two algorithms,and finally selects the traditional models K nearest neighbours(KNN),bagging(Bagging)and extreme gradient boosting tree(XGBoost)voting to generate the KBX model and train the dataset,which can effectively solve the problem of low accuracy of individual classifiers.Finally,the proposed two algorithms were extensively experimentally compared and validated using the transactions of credit card users in various European regions in September2013 as a dataset,using evaluation metric guidelines.The experimental results show that the MHA-DNN-Focal-based credit card fraud detection model has an accuracy rate of 85% and an AUC value of 99%,while the KBX-based integrated learning detection model has an accuracy rate and an AUC value of 100%,which is far better than the MHA-DNN-Focal model,proving that the integrated model has good anti-fraud results and superior performance,and better solves overfitting and other It provides a realistic basis and theoretical reference for financial institutions in the field of financial risk.