FST-GNN: A Fuzzy-Based Spatial-Temporal Graph Neural Network for Traffic Flow Prediction During Emergent Epidemics

Traffic flow prediction is very essential in traffic cognition management of modern cities, especially in the case of emergent epidemics. Due to epidemics having a huge impact on citizen travel and government control, some emergencies make traffic data increasingly dynamic, irregular, uncertain, etc., which brings a significant challenge for accurate traffic estimation or forecasting. In this study, a novel fuzzy-based spatial-temporal graph neural network approach called FST-GNN is proposed for traffic flow prediction during emergent epidemics. The model is based on emergent epidemics represented by COVID-19 and Omicron which introduces a fuzzy inference mechanism to obtain the fuzzy representation and extract features of emergent epidemic-related data such as confirmed cases, confirmed deaths, and government response indicators. And, graph neural networks are constructed to capture the dynamic spatial and temporal correlation of traffic data. At last, experiments on the two real-world datasets (bikes and taxis) in New York City are developed to show that our model achieves better effectiveness and superiority compared to other existing models.