A data-driven uncertainty quantification framework for mechanistic epidemic models
Résumé
Mechanistic epidemic models are frequently used to predict the evolution of infectious disease outbreaks. However, the uncertain nature of the model parameters (epidemiological parameters, initial conditions, etc.) and the limited horizon of predictability of this dynamic phenomenon make it essential to quantify the underlying uncertainties. In this sense, this work presents a cross-entropy approximate Bayesian computation framework for uncertainty quantification that is particularly interesting for use in epidemic models. The new methodology is tested with actual data from a COVID- 19 outbreak, presenting a great capacity to capture the variability and dynamic evolution of the disease records.