SEIR model combined with LSTM and GRU for the trend analysis of COVID-19

Abstract

Coronavirus disease (COVID-19) is an infectious disease caused by a newly discovered coronavirus, which has become a worldwide pandemic greatly impacting our daily life and work. A large number of mathematical models, including Susceptible-Exposed-Infected-Removed (SEIR) model and deep learning methods, including Long-Short-Term-Memory (LSTM) and Gated Recurrent Units (GRU), have been employed for the analysis and prediction of COVID-19. The purpose of this thesis is to analyze and predict the epidemic trend of COVID-19 in different countries by combining the SEIR model with the classic LSTM and GRU methods, and to explore the application potential of LSTM and GRU in COVID-19 epidemic trend prediction.

The core content of this thesis consists of two parts. The first part is about the learning and prediction of dynamic parameters. The parameters in the SEIR model, including infection rate and recovery rate, are constantly changing over time, and can be considered as a time series. We learn and predict the dynamic changes of these two parameters over time using LSTM and GRU and find the constantly changing reproduction rate which is closely related to them. Then, we discuss and analyze the relationship between the reproduction number and the epidemic trend of COVID-19 by simple linear fit. In the second core part, we employ LSTM, GRU and SEIR models with the dynamic parameters that were learned and predicted by LSTM and GRU to do the prediction of the epidemic trend of COVID-19 for the United States. We utilize three common error metrics, Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and r_2 score, to compare and study the results and explore the application potential of LSTM and GRU in COVID-19 prediction. Mathematical software, like Python, are used in this investigation.