Development and evaluation of frameworks for real-time bus passenger occupancy prediction

One critical aspect of bus service quality that influences riders’ attitudes is the availability of seating and/or space to board vehicles. Unfortunately, little attention has been given to short-term passenger occupancy predictions on individual buses. This research examines the use of conventional linear regression models and a machine-learning (random forest) model to predict passenger occupancies on individual buses when they arrive at future stops using data available in real-time from bus operations (e.g., Automatic Passenger Counter (APC) systems) and weather information. Overall, the linear model (LM) and the random forest (RF) model are found to provide close estimates. Three sets of models are developed in this work to model the current and future stop pairs: a next-stop-based model that only models the occupancy at the right next stop and two models that predict the occupancy at any future stop along the bus route (called OD-pair based models). The OD-pair based models are found to predict passenger occupancies more accurately at downstream stops, regardless of whether the LM or RF is used. Examination of the transferability reveals that models can provide reliable estimates of future data when trained with historical information if demand patterns are fairly stable. These models and insights can be used by transit agencies in improving the quality and breadth of information provided to transit system users and even be integrated directly into real-time end-user feeds.