Title: Evaluating short-term forecast among different epidemiological models under a Bayesian framework

Abstract:

Forecasting of COVID-19 daily report data has been one of the several challenges posed on the governments and health sectors on a global scale. To facilitate informed public health decision making for the coming weeks, the concerned parties rely on short term daily and weekly projections generated via predictive modeling. Although, several models have been proposed in the literature, there has been a growing debate amongst researchers over model performance evaluation and finding the best model appropriate for a certain feature (cases, deaths, hospitalizations, etc.), a particular regional level (county, state, country, etc.) and more. We calibrate stochastic variants of six different growth models (i.e. logistic, generalized Logistic, Richards, generalized Richards, Bertalanffy, and Gompertz) and the basic SIR model into one flexible Bayesian modeling framework. We perform time-series cross-validation to compare prediction error metrics of the methods considered. In terms of regions, we consider the 50 states of US and Washington DC for a state-level analysis and the top 20 countries in terms of case counts for a country-level analysis. After fitting the models, we visualize the mean absolute percentage error (MAPE) and its symmetric version of the model forecasts. In general, as the models learned more and more data, the predictive performance improved drastically considering all the regions. Finally, to find the best model for a certain state or country, we average the prediction error metric over the entire period and compare the results. In conclusion, it was noted that none of the models proved to be golden standards across all the regions in their entirety.