A Graph Neural and BiLSTM Hybrid Network for Transactional Fraud Detection in Digital Payment Gateways
Judy Simon, Nellore Kapileswar
Abstract
Credit card fraud continues to be a widespread and expensive problem for the financial sector, impacting a million consumers and enterprises globally. Identifying fraudulent transactions is essential for reducing financial losses and maintaining the security of payment systems. Today, hackers and other non-legitimate users would commit fraud, which would hurt some people and ruin their identity. The transaction rate fraud, card fraud (when the card isn't present), phishing attacks, and account takeover are all common types of fraud. The hybrid approach is needed and developed to stop losses and keep them to a minimum. To address privacy and security problems, a combination of two effective algorithms known as Graph Neural Networks (GNNs) and Bidirectional Long Short-Term Memory (BiLSTM) are utilized that improves the overall performance and accuracy of the proposed model. This technique makes use of GNN to capture the preexisting relationships that are dependent on the user, device, and locations in order to turn the transactional information into a graphical framework. The structural characteristics of latent nodes and their edge representations can be learned and modeled using this GNN. After that, they are passed on to the BiLSTM, which records the patterns of time and the sequential transactions that are dependent on timing. With a significantly greater accuracy percentage of 98.9% compared to the traditional models used for the same task, the suggested method is demonstrating its efficacy in identifying behavioral anomalies and collusive fraud.