A Novel Scientific Machine Learning Method for Epidemiological Modelling

In this study, we propose a novel epidemiological framework, the SEIHQR model, which integrates Scientific Machine Learning (SciML) with interpretable compartment-based modeling. We apply two pillars of Scientific Machine Learning (SciML) - neural ordinary differential equations (neural ODEs) and universal differential equations (UDEs) - to model this system of ordinary differential equations. Although traditional epidemiological modeling relies on agent-based or compartmental methods, our approach uses SciML to retain the interpretability of compartmental models while incorporating the expressive power of neural networks, similar to the complexity achievable with agent-based models. Our findings demonstrate that for this framework, neural ODEs yield poor performance and lack interpretability. In contrast, UDEs exhibit superior performance and, through symbolic recovery techniques, achieve full interpretability. This work opens avenues for the broader application of SciML frameworks in epidemiology, particularly for forecasting and monitoring disease spread.