Harnessing multi-output machine learning approach and dynamical observables from network structure to optimize COVID-19 intervention strategies.

Abstract

The coronavirus disease 2019 (COVID-19) pandemic has necessitated the development of accurate models to predict disease dynamics and guide public health interventions. This study leverages the COVASIM agent-based model to simulate 1331 scenarios of COVID-19 transmission across various social settings, focusing on the school, community, and work contact layers. We extracted complex network measures from these simulations and applied deep learning algorithms to predict key epidemiological outcomes, such as infected, severe, and critical cases. Our approach achieved an $R^2$ value exceeding 95%, demonstrating the model's robust predictive capability. Additionally, we identified optimal intervention strategies using spline interpolation, revealing the critical roles of community and workplace interventions in minimizing the pandemic's impact. The findings underscore the value of integrating network analytics with deep learning to streamline epidemic modeling, reduce computational costs, and enhance public health decision-making. This research offers a novel framework for effectively managing infectious disease outbreaks through targeted, data-driven interventions.