Novel Public Transport Prediction Systems with Dynamic Statistical Attention Technique

Abstract

A novel machine learning (ML) framework for accurate real-time bus arrival time prediction integrates the ML model with a lightweight Dynamic Statistical Attention (DSA) technique. Real-time GPS data from Macao bus routes, synchronized with weather records, constitute the primary dataset. The DSA technique blends ML predictions with historical average arrival times to mitigate random fluctuations and improve result stability. Experimental results indicate that the proposed models, particularly KNN combined with DSA, deliver superior performance through lower mean absolute error, root mean square error, and mean absolute percentage error relative to previous hybrid neural network approaches. This framework provides a computationally efficient solution suitable for deployment in resource-constrained smart city public transport systems while sustaining high prediction accuracy under varying traffic and weather conditions.