Computational Models for COVID-19 Dynamics Prediction

In a viral pandemic, predicting the number of infected per day and the total number of cases in each wave of possible variants is intended to aid decision-making in real public health practice. This paper compares the efficiency of three very simple models in predicting the behavior of COVID-19 in Spain during the first waves. The Verhulst, Gompertz and SIR models are used to predict pandemic behavior using past daily cases as observed data. The parameters of each model are identified at each wave by solving the corresponding inverse problem through a member of the PSO family and then their posterior distribution is calculated using the Metropolis-Hastings algorithm to compare the robustness of each predictive model. It can be concluded that all these models are incomplete without the corresponding parameter uncertainty analysis. In these cases, the comparison of the posterior prediction with respect to the predictive model used shows that this work can be used for real-life decision making.