Enhancing bus arrival time predictions in transit networks through spatio-temporal forecasting

Abstract

Urban public transport is a key infrastructural element that tends to suffer from the variability of bus arrival times, leading to long wait times and increased passenger dissatisfaction. This research presents a sophisticated prediction framework that leverages the synergy of Graph Neural Networks (GNNs) and Transformer-based models to overcome these limitations. Utilizing comprehensive datasets from the New York City Metropolitan Transportation Authority (MTA), the model captures both the temporal dynamics of bus movements and the spatial interdependencies among stops. The integrated model not only performs better than traditional models that tend to study routes independently, but also provides real-time accurate forecasting. Experimental results demonstrate significant improvements in predictive accuracy over established baselines, validating the model’s effectiveness. These promising outcomes highlight the potential of advanced machine learning techniques to revolutionize urban transit management and foster improved mobility.