Graph Neural Networks for Fraud Detection in E-commerce Transactions

E-commerce platforms face increasing fraud, resulting in significant financial losses, and traditional fraud detection approaches often fail to capture complex patterns. This paper proposes a Graph Neural Network approach to enhance fraud detection by modeling e-commerce data as a graph, where nodes represent consumers, sellers, and transactions, and edges represent interactions. By utilizing a multi-layer Graph Convolutional Network and message-passing algorithms, the GNN captures intricate relationships, uncovering hidden fraud patterns through local and global node interactions. Trained on a large-scale, real-world dataset, this method significantly outperforms traditional approaches like logistic regression and decision trees, demonstrating its superior ability to detect sophisticated fraud in e-commerce environments.