Learning and inferring transportation routines
Artificial Intelligence
Similarity-based prediction of travel times for vehicles traveling on known routes
Proceedings of the 16th ACM SIGSPATIAL international conference on Advances in geographic information systems
Assessing the predictability of scheduled-vehicle travel times
Proceedings of the 17th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems
Incremental learning of statistical motion patterns with growing hidden Markov models
IEEE Transactions on Intelligent Transportation Systems
Predestination: inferring destinations from partial trajectories
UbiComp'06 Proceedings of the 8th international conference on Ubiquitous Computing
| Hi-index | 0.00 |
Automatic vehicle location (AVL) systems such as those installed on public bus services produce a large amount of route specific GPS data. Many cities, Dublin Ireland being among the newest, are using this data to display time of arrival (TOA) predictions on smart screens at every bus stop on their transportation network. This paper analyses the TOA predictability of buses travelling on Dublin Bus's longest and most frequented route 46A. We propose a pattern-based route time prediction system that uses a nearest neighbour approach to combine sub-tracks of historical GPS probe data to produce prediction of time of arrival. We show experimentally that there is an optimal number of neighbours k which changes as a function of the distance travelled. We also show that long term predictions are very uncertain and we make the recommendation that these predictions be conveyed to a user as an upper/lower bound pair. We show that meaningful predictions only emerge when the bus is within 6km from its predicted bus stop.