Predicting Traffic Anomalies and Collision Risks in V2X Systems: A Deep Learning Approach Using LSTM and GNN

Vehicle-to-Everything (V2X) communication has transformed intelligent transportation systems (ITS) by enabling real-time data exchange for enhanced road safety and efficiency. However, current V2X-based safety mechanisms remain largely reactive, limiting their ability to prevent accidents in dynamic traffic environments. This paper proposes a hybrid deep learning framework that integrates Long Short-Term Memory (LSTM) networks and Graph Neural Networks (GNNs) to predict traffic anomalies and collision risks in V2X systems. The LSTM component captures temporal dependencies in vehicle behavior, while the GNN module models spatial interactions among vehicles within road networks. The fusion of these models enhances predictive accuracy, enabling proactive risk assessment and early warning generation. Experimental validation using real-world trajectory data shows that the fused model provides a balanced combination of temporal and spatial evidence; its errors are comparable to the standalone models, with the best metric depending on the selected fusion weight. The results indicate that spatiotemporal learning is a promising basis for V2X safety applications, paving the way for real-time deployment in autonomous and connected vehicle environments.