A Reinforcement Learning Based Decision Support Tool for Epidemic Control: Validation Study for COVID-19

Epidemics such as COVID-19 present a substantial menace to public health and global economies. While the problem of epidemic forecasting has been thoroughly investigated in the literature, there is limited work studying the problem of optimal epidemic control. In the present paper, we introduce a novel epidemiological model (EM) that is inherently suitable for analyzing different control policies. We validated the potential of the developed EM in modeling the evolution of COVID-19 infections with a mean Pearson correlation of 0.609 CI 0.525-0.690 and P-value < 0.001. To automate the process of analyzing control policies and finding the optimal one, we adapted the developed EM to the reinforcement learning (RL) setting and ran several experiments. The results of this work show that the problem of optimal epidemic control can be significantly difficult for governments and policymakers, especially if faced with several constraints at once, hence, the need for such machine learning-based decision support tools. Moreover, it demonstrated the potential of deep RL in addressing such real-world problems.