Predicción de la incidencia epidemiológica del dengue en Colombia utilizando una Red Neuronal Recurrente a partir de información captada por sensores remotos

Abstract

Dengue is an endemic viral disease (confined to tropical and subtropical areas) transmitted by the Aedes aegypti mosquito and is a global health concern due to its rapid spread and the increasing severity of outbreaks. This study exploits the power of machine learning algorithms to build a robust regression model from a recurrent neural network, incorporating historical data on dengue in Colombia and environmental variables derived from remotely sensed measurements. This is intended to improve the accuracy of forecasts of dengue cases, as well as to complement the state of the art for this type of epidemiological models. A total of 4 models were created using data from different historically affected departments of Colombia (Antioquia, Norte de Santander, Santander and Valle del Cauca). The models presented an RMSE of 12, 46, 20 and 15 units and an r2 of 0.89, 0.97, 0.97 and 0.88 respectively. Comparing the RMSE with the most representative central range of the data (between the 25th and 75th percentiles), it was found that, for the best performing model, the predictions are between 11% and 30% away from the real values, which is relatively small in the context of the data. This is a promising result as accurate predictions are crucial for public health planning and response allowing targeted interventions in high-risk areas, which can reduce the spread of disease and prevent outbreaks, saving lives and reducing the burden on health systems.