Towards an uncertainty aware short-term travel time prediction using GPS bus data: Case study in Dublin

In this paper we propose and study the performances of a bus travel times prediction model using real bus location data from the city of Dublin. The proposed prediction model uses a modified version of the K-Nearest Neighbors algorithm, KNN, algorithm and exhibits a significant improvement over the baseline KNN. We also investigate the benefits of decomposing travel times in three components: running time, dwell time at bus stops and time stopped at traffic lights. We discuss that most of the uncertainty on the travel times comes from time spent at bus stops and traffic lights, and prediction of running time only is much improved due to the reduced uncertainty at bus stops and traffic lights. Finally we show the need of a prediction algorithm for time spent at bus stops and traffic lights that added to the prediction of running time would allow for an uncertainty aware travel time predictor.