Comparative Analysis of AI Techniques in a Traffic Advisory Model: GNN, FNN, and A*

I Urban traffic congestion remains a persistent challenge, demanding intelligent solutions that can analyse road conditions and recommend optimal routes in real time. This study presents a comparative analysis of three Artificial Intelligence (AI) techniques Graph Neural Networks (GNN), Feedforward Neural Networks (FNN), and the A* search algorithm applied to a traffic advisory model. Each model was evaluated based on four core metrics: prediction accuracy, route optimization score, computation time, and real-time responsiveness. The models were developed and tested using simulated urban traffic data generated in Python, with libraries such as Pandas, NumPy, Matplotlib, and Seaborn for processing, modelling, and visualization. The evaluation revealed that GNN achieved the highest performance, with a prediction accuracy of 92.4%, a route optimization score of 9.1/10, and real-time responsiveness rated as 'High', though with a moderate computation time of 190 milliseconds. FNN, while being the fastest at 85 milliseconds, offered a lower prediction accuracy of 84.7% and a route optimization score of 7.5/10, and was rated 'Medium' in responsiveness. The A* algorithm, with a prediction accuracy of 65.3% and a route score of 8.4/10, showed strong deterministic pathfinding capabilities but suffered from the highest computation time of 220 milliseconds and limited adaptability. Based on the results, Graph Neural Network is recommended for deployment in complex and dynamic traffic systems, where high prediction accuracy and adaptability are crucial. FNN may be suitable for lightweight applications with strict time constraints, while A* remains useful in less dynamic or static routing environments. Future implementations should consider integrating hybrid models that combine GNN’s learning capability with A*’s deterministic efficiency to achieve both precision and speed in real-time traffic advisory systems.