Credit Card Fraud Detection Using Neural Network Auto Encoders

Abstract

Nowadays, the Internet has become a crucial part of almost every area of our lives. Online transactions, purchases, and transactions are growing exponentially on e-commerce sites. Similarly, the number of online scammers increased, and the recent increase in credit card payments makes credit card fraud a very serious threat. Institutions lose huge sums of money every year due to fraud. As these agencies tighten their security measures, scammers are constantly looking for new ways to use their nefarious methods. Well-known online payment methods include credit and debit cards, internet banking, and mobile banking. Because of its easy usage, people tend to prefer credit card payments, and as a result, fraudulent transactions related to these types of transactions are increasing. In this article, we propose a method to detect credit card fraud transactions using a neural network autoencoder. The credit card fraud detection problem is generally considered a classification problem. This requires rebalancing any data imbalances present in the dataset. In this article, we propose to treat this problem as an outlier detection problem that does not require a solution. This article also describes various parameters to consider when training a model. We also published the experimental results and conclusions. The goal of our project is to develop a robust fraud detection system that is inherently dynamic and rapidly adapts to changing fraud patterns and requires little or no manual processing during model inference and training. The proposed system architecture provides estimates of the ROC AUC(area under the receiver operating curve) at 0.9212, 0.8976, 0.9378, 0.9489, and 0.9512 at all multiples, with a cumulative ROC AUC of 0.9314 without manual coding.