Predicting confirmed cases of various epidemics using global temporal-feature-based graph convolutional network

Forecasting the spread of infectious diseases is challenging as the propagation of an epidemic is influenced by various factors. We propose the GLObal temporal feature-based Graph Convolutional Network(GLOGCN), which focuses on the global temporal patterns present in the spread of infectious diseases. Considering the existence of anomalous data in the epidemic dataset, using global temporal patterns leads the model to grasp the overall trends of the epidemic. We conducted experiments on the United States COVID-19, Hungary chickenpox, and German tuberculosis epidemic datasets to validate the performance of GLOGCN. Both graph and non-graph models were selected as baseline models for a performance comparison. Each model was trained to predict future confirmed cases of the epidemic based on the past timesteps. The experimental results show that GLOGCN achieved better performance than other baseline models in the sense of lower accumulated error. Further, we verified the model robustness on diverse patterns in the test data. The ablation study confirmed that using global temporal features had a significant impact on the model performance. In summary, GLOGCN exhibited robust performance across every dataset used in this study by processing the global temporal patterns. Thus, GLOGCN can provide supplemental data to improve epidemic policymaking.