Modeling the COVID-19 Outbreak in Various States in India
Abhishek Bhattacharjee
University of Illinois at Urbana-Champaign
This study explores various modeling and forecasting approaches that could be used to track the disease as well as its mechanics in various regions in India. Five modeling methods were used in this study and include the Exponential Smoothing Model, ARIMA Model, LSTM Model, SIR Compartmental Model, and a Prophet Library Model. All five models were used to generate 7-day forecasts with all known data as well as validation forecasts using 70% of the available data for training and 30% for testing. The SIR Model and Prophet Model were specifically used to perform a more in-depth analysis of disease spread, tackling topics such as trend changepoints, and population adjusted spread rates. With regards to inputs, the Exponential Smoothing Model, ARIMA Model, and LSTM model all relied on a univariate approach, with case count data as the only input variable. The SIR and Prophet Models utilized a multivariate approach with the SIR Model using last known population, recovery counts and death counts as extra regressors, and the Prophet Model using Google Human Mobility data as an extra factor.