Fitting Time-varying Coefficients SEIRD Model to COVID-19 Cases in Malaysia

Abstract

This paper proposes a compartmental Susceptible-Exposed-Infected-Recovered-Death (SEIRD) model for COVID-19 cases in Malaysia. This extended model is more relevant to describe the disease transmission than the SIRD model since the exposed (E) compartment represents individuals in the disease's incubation period. The mathematical model is a system of ordinary differential equations (ODEs) with time-varying coefficients as opposed to the conventional model with constant coefficients. This time dependency treatment is necessary as the epidemiological parameters such as infection rate β, recovery rate γ, and death rate μ usually change over time. However, this feature leads to an increasing number of unknowns needed to be solved to fit the model with the actual data. Several optimization algorithms under Python’s LMfit package, such as Levenberg-Marquardt, Nelder-Mead, Trust-Region Reflective and Sequential Linear Squares Programming; are employed to estimate the related parameters, in such that the numerical solution of the ODEs will fit the data with the slightest error. Nelder-Mead outperforms the other optimization algorithm with the least error. Qualitatively, the result shows that the proportion of the quarantine rule-abiding population should be maintained up to 90% to ensure Malaysia successfully reaches disease-free or endemic equilibrium.