Graph Neural Network Model in Financial Fraud Detection

Graph neural networks (GNNs) have emerged as powerful tools in computer science for learning structured data representations, especially in domains involving graph-based relationships. In this study, we apply GNNs to financial fraud detection by modeling user transactions as graphs, where nodes represent users and edges denote transaction interactions. We construct a financial transaction network and propose an end-to-end GNN-based detection model that integrates Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs) for hierarchical feature aggregation and adaptive neighbor weighting. The model consists of a multi-layer GCN module, feature extraction layer, feature fusion layer, and a multilayer perceptron classifier. Experiments conducted on the IEEE-CIS dataset demonstrate that the proposed model significantly outperforms traditional machine learning approaches including Logistic Regression, Support Vector Machines, and Random Forests, achieving higher accuracy, precision, recall, and $\mathbf{F 1}$-score. These results confirm the capability of GNNs to learn the topological structure of transaction networks effectively, offering a promising direction for intelligent fraud detection.