A Novel Hybrid Epidemic Prediction Model Based on Cross-Modal Information

To assess the threat of the COVID-19 epidemic, forecasting the number of new cases is critical for epidemic prevention. Most of the existing studies are of limited performance as they ignore two key issues: 1) there exist complex correlations between the epidemics of different regions because of the population shift and 2) the effect of nonpharmaceutical interventions (NPIs) is sensitive to their orientations and effective durations. To this end, we propose a novel cross-modal spatial–temporal epidemic prediction (CMSTEP) model for COVID-19 new-case prediction. The proposed model consists of two newly designed modules: one is a spatial–temporal sequential prediction module that captures the trend of new cases based on the historical epidemic of the target region and its related regions, the other is an intervention effect assessment module that models the NPIs’ impact based on their orientations and effective durations. Through extensive experiments on real-world data, we prove the effectiveness of the proposed model.