An Accurate Gaussian Process-Based Early Warning System for Dengue Fever

TL;DR

This paper introduces a Bayesian non-parametric Gaussian process model for dengue fever prediction in Brazil, offering a flexible and accurate early warning system to aid health authorities and improve understanding of disease dynamics.

Contribution

It presents a novel Gaussian process-based model that outperforms standard techniques and enhances dengue outbreak prediction and analysis in Brazil.

Findings

Model outperforms previous techniques in accuracy

Provides insights into dengue dynamics through covariance analysis

Enables effective early warning in Brazilian cities

Abstract

Dengue fever is a mosquito-borne disease present in all Brazilian territory. Brazilian government, however, lacks an accurate early warning system to quickly predict future dengue outbreaks. Such system would help health authorities to plan their actions and to reduce the impact of the disease in the country. However, most attempts to model dengue fever use parametric models which enforce a specific expected behaviour and fail to capture the inherent complexity of dengue dynamics. Therefore, we propose a new Bayesian non-parametric model based on Gaussian processes to design an accurate and flexible model that outperforms previous/standard techniques and can be incorporated into an early warning system, specially at cities from Southeast and Center-West regions. The model also helps understanding dengue dynamics in Brazil through the analysis of the covariance functions generated.