Adaptive Graph Neural Network Model For Traffic Flow Prediction In Intelligent Urban Transportation Systems

Abstract

Urban traffic flow is dynamic, influenced by temporal demand, roadway interactions, incidents, and fluctuating mobility patterns, creating challenges for accurate prediction and responsive transportation management. Existing approaches often inadequately represent evolving spatial dependencies and temporal variations across interconnected road networks. The research develops a Dynamic Cat Swarm Optimizer-tuned Adaptive Graph Neural Network (DCSO-AGNN) for accurate traffic flow prediction within Intelligent Urban Transportation Systems. The Los Angeles Metropolitan Traffic (METR-LA) dataset with 207 sensors and 34,272 observations and the Bay Area Performance Measurement System (PEMS-BAY) dataset with 325 sensors and 52,116 observations, recorded at 5-minute intervals, were used for traffic-flow prediction. Data pre-processing employs Savitzky–Golay Filtering to suppress sensor noise and preserve temporal traffic patterns, followed by Robust Scaling to normalize heterogeneous traffic variables.  Feature extraction uses Principal Component Analysis (PCA) to derive representations while reducing redundancy, complemented by temporal lag construction to preserve traffic behaviour. Processed observations are organized as graph-structured inputs, and connectivity relationships form edges. AGNN learns changing spatial dependencies and propagates information across connected road segments to capture evolving traffic conditions. Dynamic Cat Swarm Optimizer (DCSO) tunes AGNN parameters to improve convergence, stability, and generalization. The integrated DCSO-AGNN predicts traffic flow by jointly learning spatial interactions and temporal patterns, supporting traffic monitoring. Overall, DCSO-AGNN achieved the lowest Mean Absolute Error (MAE) of 3.12, Root Mean Square Error (RMSE) of 5.52, and Mean Absolute Percentage Error (MAPE) of 7.94%, confirming improved traffic-flow prediction accuracy using python. The model supports traffic forecasting and strengthens decision-making for responsive, efficient, and sustainable Intelligent Urban Transportation Systems.