Parameter Adjustment for Mechanistic Epidemiological Models of COVID-19: Controlling for the Impact of Metro Area Crowding

Abstract Timely understanding and accurate prediction of affected areas during novel disease outbreaks like COVID-19 is essential for the implementation of emergency response activities. Often, mechanistic models are used to evaluate the outcomes of these decisions. However, it can be difficult to estimate key features of the disease when an outbreak becoming a pandemic is in its early stages. Using compartmental model-controlled neural networks, this study creates transmission parameters for mechanistic models that are adjusted for these urban crowding features such as household size and public transportation use. The results showed that adjusted parameters can significantly improve the accuracy of the SIR model, thus helping guide resource allocation and make policy decisions.