Comparative analysis of implicit models for real-time short-term traffic predictions

Predicting future traffic conditions in real-time is a crucial issue for applications of intelligent transportation systems devoted to traffic management and traveller information. The increasing number of connected vehicles equipped with locating technologies provides a ubiquitous updated source of information on the whole network. This offers great opportunities for developing data-driven models that extrapolate short-term future trend directly from data without modelling traffic phenomenon explicitly. Among several different approaches to implicit modelling, machine-learning models based on a network structure are expected to be more suitable to catch traffic phenomenon because of their capability to account for spatial correlations existing between traffic measures taken on different elements of the road network. The study analyses and applies different implicit models for short-term prediction on a large road network: namely, time-dependent artificial neural networks and Bayesian networks. These models are validated and compared by exploiting a large database of link speeds recorded on the metropolitan area of Rome during seven months.