A Multi-Dimensional Telecom Fraud Detection Model Based on Graph Attention Networks

With the rapid development of 5G mobile networks, telecom fraud methods are continuously evolving, and the number of fraudsters and victims is increasing. Traditional telecom fraud detection methods struggle to effectively capture the complex relationships between nodes when dealing with high-dimensional, complex, and dynamic telecommunication data. These complex relationships can be modelled as graphs. With the emergence of Graph Neural Networks (GNNs), GNN-based telecom fraud detection methods have gradually become a research hotspot. We present a telecom fraud detection model based on Graph Attention Networks (GAT). The model leverages Graph Neural Networks advantages to enhance node feature representations by adaptively aggregating neighbouring nodes’ features. Then, our model employs a Multi-Layer Perceptron (MLP) to classify each node (i.e., user) and determine its involvement in fraudulent activities. In constructing the model, we incorporate multi-dimensional data such as phone numbers, call records (e.g., participants, duration, type, time, location), billing information, devices used, and user IDs to form a graph structure representing users and their communication behaviours. Experimental results demonstrate that our model significantly improves detection accuracy and robustness, particularly in handling the challenges of multi-dimensional data.