On Application of Regime-Switching Models for Short-Term Traffic Flow Forecasting

This paper contributes to the identification of spatial dependency regimes in urban traffic flows. Importance of traffic flow regimes for forecasting and presence of spatial relationships between road network nodes are widely acknowledged both in traffic flow theory and empirical studies. In this research, we join these concepts and made the first steps to analysis of different regimes of spatial dependency in a traffic flow. Modern Markov-switching autoregressive distributed lag models are utilized and allowed to analyse the model structure in different traffic flow regimes. On the base of the models, we made a conclusion about the importance of traffic flow regimes for identification of a structure of spatial dependencies. The proposed approach is illustrated for real-world traffic flow data.