Deep learning based traffic flow prediction on the real-world road network

Abstract

Traffic Flow prediction is a very important part of managing traffic flows on the road network. It plays a very important role in avoiding traffic congestion and helps in building a safer and more efficient transportation network. Traffic flow prediction is very complex as it involves dealing with traffic flow patterns that are in the form of spatiotemporal data and it also involves considering important factors such as weather patterns and the large-scale nature of the road networks. With the advancement of deep learning algorithms, dealing with spatiotemporal data has become relatively simpler. This paper proposes using a combination of deep learning algorithms, GCN, and Transformer for predicting traffic flow. To evaluate these algorithms, datasets containing information about the distance traveled by vehicles at a particular time and location have been used. Experiments results are compared with some baseline models such as LSTM, GRU and GCN.