Uncertainty in urban mobility: Predicting waiting times for shared bicycles and parking lots

Building efficient and sustainable transportation systems is a key challenge for accommodating the fast-increasing population living in cities. Lack of efficiency in transportation networks typically arises from uncertainty, e.g., about the availability of resources (such as parking lots or bicycles in bike sharing systems), or the exogenous factors affecting their demand (such as weather or the time of the day). In this paper, we present a class of algorithms which use Generalized Additive Models (GAMs) for demand and availability prediction on various time scales. In contrast to existing methods, exogenous effects can be explicitly factored into the models, resulting in significant gains in terms of prediction accuracy. Another advantage of our approach is that it estimates the distribution of the waiting time for the next available bike/parking lot if the current availability is zero. We showcase how this additional information can be used as part of personal uncertainty-aware journey planners which allow users to choose from multiple routes according to their time constraints.