Optimizing the SEIRD Model for COVID-19 in Malaysia Using Pymoo Framework

Abstract

Recently, two specialized compartmental epidemiological models; the modified SIRD and SEIRD were developed to study COVID-19 in Malaysia, with their validity tested by fitting them to actual data. In this case, the optimization problem has a single objective: minimizing the least square error between the numerical solution and real-world data, without any imposed constraints. However, the introduction of time-dependent coefficients in both models increases the number of optimization variables. To solve this, the Nelder-Mead and Pattern Search algorithms were recommended. While Nelder-Mead is widely available in Python optimization libraries, Pattern Search is less common, which motivated the choice of the Pymoo multi-objective optimization framework for this study. The fitting results show that the absolute error metrics improve highest by about 40% compared to the results obtained from other optimization packages in previous studies. Furthermore, as an extension of prior research, we incorporate the computation of the dynamic basic reproduction number and its sensitivity analysis, confirming the effectiveness of the movement control order in controlling the disease during the pre-vaccination phase.