Abstract:
|
Although mapping services (like Bing and Google Maps) provide predictions of travel time for arbitrary routes in a road network, there can be uncertainty in those predictions due for example to unknown timing of traffic signals and unpredictable traffic congestion. Predictions of the probability distribution of travel time account for the presence of these unmeasured conditions, and can be used to report travel time reliability to a user, to provide risk-averse route recommendations, and as a component of fleet vehicle decision support systems. I will present methods to predict the distribution of travel time on an arbitrary route in a road network at a given time, by accurately accounting for dependencies in travel time across road segments. Our methods are applied to large volumes of mobile phone GPS data from the Seattle metropolitan region. Efficient computation is achieved by maximum a posteriori estimation via Expectation Conditional Maximization, providing a procedure with closed-form updates in each iteration. We argue that our method is likely to be computationally feasible for the continental-scale road networks and high-volume GPS data of modern mapping services.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.