Online payment fraud continues to cause significant financial losses for businesses and consumers. Traditional machine learning approaches for fraud detection typically treat each transaction as an independent data point, ignoring relationships among users, devices, payment cards, and transaction patterns. However, fraudulent activities often exhibit complex relational structures that can be better captured through graph-based representations.

This project proposes the use of Graph Neural Networks (GNNs) to model and exploit the interconnected nature of transactions in the IEEE-CIS Fraud Detection dataset. By representing entities and transactions as nodes and their interactions as edges, the proposed approach aims to improve fraud detection performance compared to conventional machine learning models.

Dataset: IEEE-CIS Fraud Detection dataset (https://www.kaggle.com/competitions/ieee-fraud-detection/overview)