Relation between Proximity of Streets in Urban Network and Parameters of Neural Network for Traffic Volume Prediction

Deep learning has become very popular as a method to predict short-term traffic volumes on road networks, especially highway networks, based on real-time observation. Various studies have confirmed that the performance of deep learning in predicting traffic volumes is better than that of previous machine learning models and statistical models. Although it is natural to consider that the traffic conditions on road networks are highly dependent on network structures such as the connection relationship between roads, to date it is not clear whether the estimated parameters of neural networks are related to the proximities of roads in networks. This study was conducted with the objective of predicting traffic volumes in urban street networks, which are more complex than highway networks, and investigated the relation between proximity of streets and estimated weight parameters of a neural network. The results obtained confirm that the proximity of streets is significant in traffic volume prediction, although some streets have a strong relation with distant streets.