Prediction of Journey Destination in Urban Public Transport

Abstract

In the last decade, public transportation providers have focused on improving infrastructure efficiency as well as providing travellers with relevant information. Ubiquitous environments have enabled traveller information systems to collect detailed transport data and provide information. In this context, journey prediction becomes a pivotal component to anticipate and deliver relevant information to travellers. Thus, in this work, to achieve this goal, three steps were defined: (i) firstly, data from smart cards were collected from the public transport network in Porto, Portugal; (ii) secondly, four different traveller groups were defined, considering their travel patterns; (iii) finally, decision trees (J48), Naive Bayes (NB), and the Top-K algorithm (Top-K) were applied. The results show that the methods perform similarly overall, but are better suited for certain scenarios. Journey prediction varies according to several factors, including the level of past data, day of the week and mobility spatiotemporal patterns.