Fraud Detection in Financial Services Using Siamese Network

Abstract

In financial services, it is important to detect criminal transactions such as fraudulent money transfers. However, fraud detection in conventional services is often rule-based, which poses challenges in terms of detection accuracy and operational costs. In addition, research on fraud detection using machine learning has been conducted, but one of the problems is that learning is not easy due to the very high imbalance of the data. In this paper, we propose a fraud detection model based on Siamese Network. Siamese networks are often used for image recognition tasks, in which the model is used to determine whether two images are in the same class. An important advantage for fraud detection tasks is that having two inputs makes it easier to deal with imbalances. We have partially modified this network and applied it to fraud detection. The data used in the experiment are dummy data that replicate the characteristics of actual financial service data. We apply the proposed model to the dummy data and show that the accuracy of the model is improved over the previous method.