Predicting traffic intensity in the urban area of Madrid: Integrating route network topology into a machine-learning model

Abstract

In addition to temporal effects, spatial effects must be considered in models of traffic intensity. Temporal effects (temporal autocorrelation) and spatial proximity shape observed traffic intensity, with repetitive patterns of peak and trough hours on weekdays and simultaneous peaks in traffic intensity picked up by traffic sensors along roadways of similar capacity. In this article we explore the ability of a neural network (NN) model to replicate both patterns, with the explicit inclusion of the route network’s characteristics. Combining information on optimal routes between different origins and destinations in the Madrid metropolitan area with the localization of traffic sensors, we estimate the route use intensities that mimic the observed traffic patterns as those that can most accurately reproduce the traffic-intensity values recorded by the sensors throughout the urban area. We conclude that this modeling approach can in fact mimic the distribution of the main origins and destinations of displacements in the metropolitan area of Madrid, reproducing the spatial and temporal patterns observed in traffic intensity. In addition, we conduct an analysis of anomalies to test an eventual application of the proposed method to determine the effects on the traffic system of critical events, such as the implementation of the Madrid Central Low Emission Zone.