Abstract
Public road authorities and private mobility service providers needinformation derived from the current and predicted traffic states to act uponthe daily urban system and its spatial and temporal dynamics. In this research,a real-time parking area state (occupancy, in- and outflux) prediction model(up to 60 minutes ahead) has been developed using publicly available historicand real time data sources. Based on a case study in a real-life scenario inthe city of Arnhem, a Neural Network-based approach outperforms a RandomForest-based one on all assessed performance measures, although the differencesare small. Both are outperforming a naive seasonal random walk model. Althoughthe performance degrades with increasing prediction horizon, the model shows aperformance gain of over 150% at a prediction horizon of 60 minutes comparedwith the naive model. Furthermore, it is shown that predicting the in- andoutflux is a far more difficult task (i.e. performance gains of 30%) whichneeds more training data, not based exclusively on occupancy rate. However, theperformance of predicting in- and outflux is less sensitive to the predictionhorizon. In addition, it is shown that real-time information of currentoccupancy rate is the independent variable with the highest contribution to theperformance, although time, traffic flow and weather variables also deliver asignificant contribution. During real-time deployment, the model performs threetimes better than the naive model on average. As a result, it can providevaluable information for proactive traffic management as well as mobilityservice providers.