A Framework for Real-time Bus Travel Time Prediction with Reliability Sensitivity

Surface transit systems face many challenges in their operation, which can result in poor schedule adherence. Passenger information systems that predict arrival times using real-time data from location tracking devices on transit vehicles can reduce the inconvenience caused by schedule deviations. However, real-time location data streams are often noisy, incomplete or inaccurate, requiring special consideration in real-time prediction applications. Additionally, transit may operate in conditions of high or low service reliability, which may require different treatments. This study proposes and tests a real-time prediction framework that uses measures of both service reliability and data reliability/quality to characterize conditions while making predictions sequentially and leveraging a variety of Machine Learning methods. The tests performed show that the proposed framework offers distinct advantages, while additional investigations carried out as part of the study highlight the framework's limitations and potential for future research.M.A.S