ON THE BAYESIAN ZERO-INFLATED SPATIO-TEMPORAL MODELLING OF DENGUE HEMORRHAGIC FEVER

Abstract

This study focuses on improving the convergence rates of parameter estimates for two Bayesian spatio-temporal models, namely the Bayesian zero-inflated Poisson spatio-temporal (BZIP S-T) distribution and the Bayesian zero-inflated negative binomial spatio-temporal (BZINB S-T) distribution, employed for modeling dengue hemorrhagic fever (DHF) data in the Caraga region, Philippines. The predictive performance of these models, incorporating meteorological factors such as rainfall and population density, is enhanced through the implementation of an overrelaxation algorithm designed to expedite convergence. Markov chain Monte Carlo (MCMC) techniques, specifically utilizing the full conditional distribution, are utilized for parameter estimation.

Our findings reveal that the application of the overrelaxation algorithm yields significant improvements in the convergence rates of parameter estimates, with acceleration percentages of up to 67% and 40% observed for the BZIP S-T and BZINB S-T models, respectively. Notably, both the models identify rainfall and population density as statistically significant predictors for DHF case predictions in the Caraga region, Philippines. While the BZINB S-T model exhibits the smallest deviance, both the models prove to be valuable tools for predicting DHF cases in the region, contributing to the advancement of epidemiological research and public health planning.

Received: April 4, 2023
Revised: September 13, 2023
Accepted: September 20, 2023