Inferring destination from mobility data
Naeem, Ali A.
Brown, Kenneth N.
CEUR Workshop Proceedings (CEUR-WS.org)
Destination prediction in a moving vehicle has several applications such as alternative route recommendations even in cases where the driver has not entered their destination into the system. In this paper a hierarchical approach to destination prediction is presented. A Discrete Time Markov Chain model is used to make an initial prediction of a general region the vehicle might be travelling to. Following that a more complex Bayesian Inference Model is used to make a fine grained prediction within that destination region. The model is tested on a dataset of 442 taxis operating in Porto, Portugal. Experiments are run on two maps. One is a smaller map concentrating specificially on trips within the Porto city centre and surrounding areas. The second map covers a much larger area going as far as Lisbon. We achieve predictions for Porto with average distance error of less than 0.6 km from early on in the trip and less than 1.6 km dropping to less than 1 km for the wider area.
Markov chain model , Bayesian inference model , Destination prediction , Taxis
Naeem, A. A. and Brown, K. N. (2017) 'Inferring destination from mobility data', Proceedings of the 25th Irish Conference on Artificial Intelligence and Cognitive Science, Dublin Institute of Technology, 7 - 8 December, pp. 153-165
© 2017, the Authors. Copying permitted for private and academic purposes.