The safe and efficient transportation of hazardous materials remains a critical challenge in Intelligent Transportation Systems (ITS), demanding real-time optimization of cost, risk, and duration. This study presents a novel ITS framework that leverages Graph Neural Networks (GNNs) and Federated Learning to model transportation networks as weighted, directed graphs. By encoding real-time traffic, environmental, and infrastructure risk data into the graph’s nodes and edges, the system dynamically predicts optimal routing paths for dangerous goods. A Python-based framework was developed to construct, train, and evaluate the model across varying node sizes (7 to 100), measuring performance with accuracy, precision, recall, F1-score, and loss. The model outperformed traditional routing algorithms with predictive accuracy up to 90% in both simulation and real-world scenarios. These findings highlight the potential of GNN-driven ITS for enhancing safety and efficiency in smart city logistics.