An AI-Enabled Agent-Based Simulation Platform for Studying COVID-19 Pandemic

Abstract

Understanding outbreak dynamics is essential for designing effective control measures. We developed an agent-based model to examine how changes in epidemiological and intervention parameters affect infection progression in a synthetic population. The model incorporates individual demographic characteristics, including age, sex, and working status, as well as the number and location of infection epicentres, diagnostic sensitivity, the proportion of asymptomatic infections, and the timing and duration of lockdowns. By tracking each individual, the simulator characterizes infection progression through a community over time. In a closed population of 10000 people, cases peaked around the sixth week and declined by approximately the fifteenth week in the absence of lockdown. When primary cases were introduced within densely populated clusters, cases peaked earlier and declined more slowly. Lockdowns delayed and reduced the infection peak, whereas lower diagnostic sensitivity increased cases and deaths. The number of cases decreased as the proportion of asymptomatic infections increased under the model's assumptions. The model produces reproducible estimates under realistic parameter settings and can accommodate factors such as infectivity period, testing yield, socioeconomic status, daily travel, awareness, population density, and social distancing. It can also be adapted to infections with similar transmission dynamics. The model is available as an open, interactive web application that enables users without programming experience to design scenarios and examine outbreak dynamics in real time. Beyond forecasting, the simulator provides a reusable in-silico environment, or digital twin, for synthetic-data generation and AI-assisted optimization of intervention policies.

9 pages, 4 figures, 1 table