STAMP: An Approach to ETA Prediction by Spatio-temporal Discretization and Machine Learning

In this paper we present a novel approach called STAMP for predicting the ETA of a query trip under the condition that the route to be taken is unknown. Unlike most of the existing approaches which compute a plausible route for the query trip and predict ETA based on the computed route, STAMP does not need to compute any route at the time of prediction. Instead, STAMP partitions a road network offline, into ETA homogeneous zones, and pre-computes k-shortest paths between each zone pair to accommodate various route choices of travelers. It then builds a machine learning model using expressive features created based on the k-shortest paths. These features capture the road characteristics and traffic patterns between a zone pair, without concerning the exact origin/destination location within a zone. We compare STAMP with two existing approaches using industrial scale real-world data. The results show that STAMP provides better prediction accuracy and is able to generalize across space (unseen location pairs) and time (unseen departure times).