Reaserch On Credit Card Fraud Unbalanced Classification Based On Generative Adversarial Nets

Credit card fraud detection is an important part of bank risk management.Along with the popular of credit cards,the amount of transaction data is rising as well.Among them,the majority is legal data,and the fraudulent transaction only takes a minor part.However,the loss bring by this is dramatically huge.how to detection the fraud is a critical issue facing the bank.Credit card fraud is actually an imbalanced classification issue and small class samples are often the focus of our attention.That is to say,how to increase the recognition rate of small class samples is the key to solve the problem of unbalanced classification.This paper analyzes and studies the unbalanced classification,and the summarizes the advantages and disadvantages in unbalanced classification.For the imbalanced data classification,traditional single classifiers may not efficiently predict the class of data.Wasserstein Generative Adversarial Nets-Adaptive Boosting(WGANBoost)as a novel binary-class imbalanced data classification algorithm was proposed focused on generative adversarial nets and ensemble learning.Firstly,wasserstein generative adversarial nets was adopted to get a generative model.The generative model produced new minority class samples to reduce imbalance ratio.The new minority class samples was brought into Adaptive Boosting learning framework to update weights and improve Adaptive Boosting algorithm.This improvements improved classification performance of Adaptive Boosting.The algorithm used a Area Under the Carve(AUC)and F measure to evaluate the performance of classifier when dealing with imbalanced classification problems.The experimental results on validation data set illustrate that,comparted with other sampling methods,the performance of WGAN is better;comparted with other classification methods,the performance of WGANBoost is better.Finally,the methods proposed this paper are applied to analyze credit card default data and real data results confirmed that the methods proposed can effectively improve the classification performance of imbalanced data.