A Bayesian - Deep Learning model for estimating Covid-19 evolution in Spain

TL;DR

This paper introduces a semi-parametric Bayesian-Deep Learning approach combining sequence analysis and Bayesian modeling to estimate and predict Covid-19 evolution across Spanish regions, accounting for uncertainty.

Contribution

It presents a novel hybrid model that integrates deep learning with Bayesian inference to improve Covid-19 case count predictions and uncertainty quantification.

Findings

Accurately models Covid-19 incidence sequences across regions.

Provides reliable uncertainty estimates for future case predictions.

Enables scenario analysis for Covid-19 evolution in Spain.

Abstract

This work proposes a semi-parametric approach to estimate Covid-19 (SARS-CoV-2) evolution in Spain. Considering the sequences of 14 days cumulative incidence of all Spanish regions, it combines modern Deep Learning (DL) techniques for analyzing sequences with the usual Bayesian Poisson-Gamma model for counts. DL model provides a suitable description of observed sequences but no reliable uncertainty quantification around it can be obtained. To overcome this we use the prediction from DL as an expert elicitation of the expected number of counts along with their uncertainty and thus obtaining the posterior predictive distribution of counts in an orthodox Bayesian analysis using the well known Poisson-Gamma model. The overall resulting model allows us to either predict the future evolution of the sequences on all regions, as well as, estimating the consequences of eventual scenarios.