A Hybrid EBM–ABM Framework for COVID-19 Modeling via Sequential SEIRD Calibration with Particle Swarm Optimization in Mexican Cities

The efficiency of epidemiological models is based on two characteristics: speed and accuracy. While Equation-Based Modeling (EBM) enables agile calculations, its approximations lack precision because they exclude population heterogeneity. On the other hand, Agent-Based Modeling (ABM) integrates individuality into the model, but it requires considerable processing time to obtain results. In this work, a sequential strategy is developed to generate ABM that preserves population heterogeneity while reducing processing time. This model is obtained through an optimization process that identifies the set of parameters whose approximation to the pandemic’s real data minimizes error. The process begins with Particle Swarm Optimization (PSO), whose cost function is implemented via EBM, enabling comparison of the population projections against the real data recorded by health institutions in three major Mexican cities. The resulting parameters are transferred to an ABM modeled in NetLogo and validated using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the Coefficient of Determination (R2). The results show that calibrated parameters during the optimization stage, when applied in ABM simulations, generate epidemiological scenarios that reflect the pandemic’s actual behavior. These findings indicate that EBM-based calibration supports ABM experimentation while reducing the computational cost of agent-level optimization.