Parameter calibration and uncertainty quantification in an SEIR-type COVID-19 model using approximate Bayesian computation
Résumé
COVID-2019 disease is affecting all the world, and it seems that we will have to live with it for a long time. To better understand its dynamics and to help in predicting possible scenarios, mathematical models have been proposed to analyze the dynamics of COVID-19. If a model is aimed to be part of a digital twin that will be used to support decisions, one important step is to calibrate its parameters with data from a specific region. In this paper, a SEIRD epidemic model is employed in this endeavor. This model considers the evolution of the proportion of susceptibles, exposed, infected, and recovered individuals, as well as deaths. The aim of this work is twofold. The first one is to calibrate the SEIRD epistemic model using real data from the local COVID-19 outbreak, to use as a decision tool in the location of interest. For this purpose, a relatively novel Monte Carlo-like metaheuristic dubbed the cross-entropy method is applied in the model calibration process. It is a general stochastic search algorithm that provides an adaptive way to find the optimal importance sampling distribution. The method is based on an iterative procedure that uses the results obtained in one interaction to produce better results in the next interaction. The second goal of the present investigation is to construct a consistent probabilistic model considering some parameters as random variables and to quantify the impact of parameter uncertainty in the epidemiological predictions. The proposed framework is illustrated for several countries and Brazilian states and cities.
Origine : Fichiers produits par l'(les) auteur(s)