Route Optimization and Recommendation System Using Graph Neural Networks: A Case Study of Port Harcourt Urban Road Corridors

Urban traffic congestion in Port Harcourt, Rivers State, Nigeria, remains a critical infrastructural challenge, with average peak-hour vehicle speeds on key corridors falling below 5 km/h at several major intersections. This study presents a Graph Neural Network-based Route Optimization and Recommendation System (GNN-RORS) applied to four interconnected urban road corridors such as Ikwerre Road, Government Reserved Area (GRA), Ada George Road, and Iwofe Road. The road network is formalized as a directed weighted graph where intersections are nodes and road segments are dynamic edges. A hybrid architecture combining Graph Attention Networks (GAT) for spatial encoding with Long Short-Term Memory (LSTM) networks for temporal encoding is proposed to simultaneously capture cross-corridor traffic dependencies and time-varying congestion patterns. Furthermore, traffic data collected across 28 days at seven monitored intersections yield a clean dataset of 7,627 records for model training, validation, and testing. The proposed model achieves a Mean Absolute Error (MAE) of 2.34 vehicles/min, outperforming five established baselines including DCRNN (MAE 3.56), STGCN (MAE 3.12), and standalone LSTM (MAE 4.81). Route recommendations generated from predicted edge-weight graphs reduce average journey times by 31.2% across five major origin-destination pairs. The congestion heatmap analysis confirms that GNN-based re-routing redistributes traffic load more evenly across corridors, reducing peak-hour congestion indices at the most affected intersections by up to 37%. These findings demonstrate that GNN-based optimization is both technically feasible and practically impactful for urban transport management in Sub-Saharan African cities.