Modeling the impacts of governmental and human responses on COVID-19 spread using statistical machine learning

Abstract

Understanding the impacts of governmental and human responses on the pandemic control is imperative for forecasting pandemic spread under various responsive scenarios and guiding localized interventions before pharmaceutical interventions are available. This study analyzed multiple data sets, including social media, mobility, policy evaluations, and COVID-19 infection reports, to delineate the interactions between governmental and human responses and COVID-19 spread in the United States in 2020 when vaccinations were unavailable. The contributions are (1) uncovering the spatiotemporal variations in governmental and human responses during COVID-19; (2) developing a statistical machine learning algorithm that incorporates spatiotemporal dependencies and temporal lag effects to model the relationships between governmental and human responses and the pandemic spread; (3) dissecting the impacts of human responses on the pandemic across space and time. Results reveal that the determinants of COVID-19 health impacts transitioned from human mobility during the initial outbreak phase to both human mobility and stay-at-home policies during the rapid spread phase, and ultimately to the compound of human mobility, stay-at-home policies and the public awareness in the full-blown phase. These findings furnish guidance for policymakers in implementing adaptive and phased strategies.

Keywords

• Bayesian model
• COVID-19
• human responses
• spatiotemporal
• geospatial data