Data as a resource: real-time predictive analytics for bus bunching

This paper outlines the results of an eight week pilot program in Miami that provided a real-time system to predict bus bunching along four routes. Bunching refers to the phenomenon of two buses serving the same route arriving at a stop at the same time. Desired service headways are not maintained in such cases leading to lower levels of service for passengers on account of longer wait times and lower utilization of resources for operators. Predictions of when and where bus bunching is likely to occur in the future equip controllers with tools to regularize headways in an anticipative fashion. The ‘best-effort’ prediction system works in an entirely data-driven fashion and leverages, in part, a historical catalog of bus propagation. Applications in performance management and schedule optimization are also presented.