An Optimized Graph Neural Network Approach for Traffic Flow Prediction in Urban Transportation Systems

Traffic flow prediction is necessary for the successful functioning of the urban transport system, as it allows predicting traffic flow and managing it in advance, reducing congestion and increasing urban mobility. The paper presents an optimal Bird Swarm Graph Neural Network (BS-GNN) predictive model of traffic flow in urban road networks in both space and time. The proposed model represents intersections and road segments as graph nodes, captures spatial dependencies through graph neural network operations, and learns temporal dynamics via sequential propagation. Bird Swarm Optimization (BSO) is used to tune network parameters dynamically to achieve better prediction. Extensive simulations indicate that BS-GNN is more successful than normal GCN, GAT, and ST-GCN models, with a Mean Absolute Error (MAE) of 5.82 vehicles/hour, Root Mean Square Error (RMSE) of 9.12 vehicles/hour, Mean Absolute Percentage Error (MAPE) of 4.12, R² score of 0.962, and congestion F1-score of 0.91 at a prediction horizon of 5 min. The model also performs well, with an MAE of 6.75 vehicles/hour, an RMSE of 10.32 vehicles/hour, an MAPE of 5.21, an R² of 0.951, and an accuracy of 94.6. The model can also be easily scaled to networks of 1000 nodes, and the computation time per epoch is 28.4 s, which is impressive, as maintain reliable performance even under peak traffic, congestion, and incidents. These findings substantiate that BS-GNN is a valid, scalable, and high-accuracy solution for real-time traffic prediction and smart transportation systems.