Probabilistic Bus Delay Predictions with Bayesian Networks

Abstract

Numerous decisions of users and operators of public transport systems depend on the availability of good arrival and departure time predictions. Passengers decide on departure time, route choice, or mode choice and operators decide on schedules, timetables, rolling stock allocation, or control actions. In practice, not only the most likely value of a bus delay is of interest, but also its variability. This paper focuses on the probabilistic prediction of bus delays with realtime information. The dynamics of bus operations are modeled by a Bayesian network framework, allowing the description of the time-dependent stochastic processes of delay evolution. The model structure can capture the dependencies between bus operation, passenger ridership, and road demand. The application to urban bus lines in Zurich, Switzerland, shows an increased prediction accuracy compared with other methods. The model allows predicting the associated variability of bus delays and provides, therefore, the basis for more accurate passenger information and risk-based decisions making of operators.