Graph Convolutional Network and Attention for Traffic Prediction - A Deep Learning Approach

Background: Traffic prediction is a key component of the intelligent transportation system for researchers and practitioners. It is extremely challenging because traffic flows typically exhibit complicated patterns, complex spatio-temporal correlations, and non-linearities. Objective: Prediction of traffic density on the roads can help not only urban traffic management but also support for other road services such as path planning. Methods: In this paper, we propose an attention-based graph convolutional network (AGCN) model to solve the traffic prediction problem. The primary focus of AGCN is on temporal, daily, and weekly dependencies of traffic periodicity. To efficiently capture the dynamic geographical and temporal correlations in traffic data, an attention-based spatial-temporal mechanism is employed. Additionally, standard convolutions are employed to extract temporal data and graph convolutional networks are used to capture spatial patterns. Results: The final prediction results are generated by fusing the outputs of these components. California Transportation Agencies Performance Measurement System (CalTrans PeMS) dataset is used in this research to assess the performance. The proposed model has been validated using simulations that exhibit the viability of the method and show 4% increase in the accuracy of prediction. Conclusion: For improved route planning and to arrive at the destination in the least amount of time, an efficient traffic pre- diction model is suggested. This enhances overall transportation system efficiency and aids in traffic control.