Retrospective analysis and forecast of Covid-19 pandemic

Abstract

The COVID-19 pandemic has caused huge negative impacts all over the world. The government needs to spend overwhelming resources to decrease the spread of the virus. However, depending only on the data of the number of daily confirmed cases, society cannot respond to the current epidemic situation very well. Compared with the daily infections, the data of confirmed cases is lagging. Some patients are also not counted. More information is needed for the government to reduce the gap between reality. Math- ematical models can help people study infectious diseases. Various models have been proposed to depict the pandemic and infer more information. Guided by retrospective analysis of the past pandemic and short-term forecasts from the model, the government can allocate medical resources effectively and carry out control policies. In this thesis, we introduce and extend the compartmental model in epidemiology and a new statisti- cal model based on regression to model the pandemic dynamic. Besides epidemiological data, we also borrow information from other datasets, such as mobility data and sero- prevalence data. Our goal is to compare the result of retrospective analysis and forecast among different model setups and find an optimal method to model the pandemic with the data of COVID-19 in the U.S. as an example.