Real-Time Bus Arrival Time Prediction: A Practical Data-Driven Approach

With growing support for diversifying travel options and efforts to improve sustainability in the transportation industry, transit system growth is anticipated worldwide. With this, comes an opportunity to leverage new vehicle tracking technologies and big data to improve transit services and attract users. The demand for reliable transit systems paired with widespread access to transit data highlights a need for a model capable of accurately predicting transit travel time. Although numerous studies have explored the bus travel time prediction modelling problem, thus far no method has emerged as consistently superior, and most complex models are difficult to interpret. Previous research has focused primarily on short-term predictions and few studies have investigated the effects of policy adherence and driver behaviour on prediction accuracy. The goal of this thesis is to develop a practical data-driven model capable of accurate bus arrival time prediction that can incorporate real-time data and generate predictions for a medium-term forecast horizon, capture real-world transit conditions, and provide a high level of transparency and interpretability. Following extensive data compilation and consolidation, including the development of a novel process for combining APC and AVL data, a bus arrival time prediction model is developed. Bus arrival time is predicted at downstream stops on the same trip or the upcoming trip by combining link travel time and stop dwell time components. Each component is predicted using historical observations and real-time information while considering operational challenges and strategies. The model performs well, particularly for a medium-term forecast range. It outperforms scheduled predictions for forecast horizons exceeding 30 minutes and has a mean absolute error (MAE) below six minutes for forecast horizons up to two hours. Prediction error is generally proportional to the forecast horizon and any problematic route segments or stops are easily identifiable. Overall, the model’s transparency, interpretability, and prediction accuracy make it a practical solution for bus travel time forecasting. It enables transit agencies to identify system weaknesses and improve service reliability by adjusting services or operations and provide more accurate information to passengers. Ultimately attracting users, reducing traffic congestion, and improving the safety and sustainability of the transportation network.