Spatio-Temporal Graph Convolutional Networks for Short-Term Traffic Forecasting

Accurate and timely provided network traffic data is important for a large number of applications in traffic management, urban planningm and guidance. Traffic forecasting remains a very challenging problem since the traffic flows usually show complex non-linear traffic patterns and have spatial dependencies on the road networks. Existing methods and algorithms usually consider spatial and temporal correlations in traffic data separately. In this paper, we investigate deep convolutional neural networks on graphs to solve short-term traffic forecasting problems. The considered graph convolutional networks are able to efficiently capture spatio-temporal correlations in traffic data. Experimental results show that the considered model outperforms the baseline methods on the transportation network of the Samara city, Russia.