Identification of an Epidemiological Model to Simulate the COVID-19 Epidemic using Robust Multi-objective Optimization and Stochastic Fractal Search

TL;DR

This paper develops a robust multi-objective optimization approach using stochastic fractal search to accurately estimate COVID-19 model parameters, accounting for uncertainties and sensitivities in the inverse problem.

Contribution

It introduces a novel robust inverse problem formulation for epidemiological modeling, integrating stochastic fractal search and sensitivity analysis for improved parameter estimation.

Findings

More reliable parameter estimates for COVID-19 dynamics.

Enhanced robustness against uncertainties in model variables.

Effective identification of epidemic behavior using real data from China.

Abstract

Traditionally, the identification of parameters in the formulation and solution of inverse problems considers that models, variables and mathematical parameters are free of uncertainties. This aspect simplifies the estimation process, but does not consider the influence of relatively small changes in the design variables in terms of the objective function. In this work, the SIDR (Susceptible, Infected, Dead and Recovered) model is used to simulate the dynamic behavior of the novel coronavirus disease (COVID-19), and its parameters are estimated by formulating a robust inverse problem, that is, considering the sensitivity of design variables. For this purpose, a robust multi-objective optimization problem is formulated, considering the minimization of uncertainties associated to the estimation process and the maximization of the robustness parameter. To solve this problem, the Multi-objective Stochastic Fractal Search algorithm is associated with the Effective Mean concept for the evaluation of robustness. The results obtained considering real data of the epidemic in China demonstrate that the evaluation of the sensitivity of the design variables can provide more reliable results.