Modeling Arbovirus Outbreak Forecasting Based on Case Concentration and Climatic Factors Using an Echo State Neural Network
摘要
Arboviruses represent a serious threat to public health, causing suffering and significant impact on health systems, especially in the public sector. The recent outbreaks of these diseases have highlighted the lag and fragility of existing prevention and control measures. Failures in control strategies for the transmitting vector, as well as structural issues such as lack of urban planning and lack of basic sanitation, exacerbated by unfavorable socioeconomic conditions, contribute to the worsening of the disease. This article aims to predict the incidence of cases and epidemics at specific intervals and proposes the use of Machine Learning techniques to predict outbreaks and increases in the incidence of arboviruses, incorporating climate data. Using a database with geographic information, maps of the spatial distribution of arbovirus cases, rainfall, temperature and wind speed for the city of Recife. The data were obtained from the Portal de Dados abertos da Prefeitura do Recife, SIGRH and APAC for the years 2013 to 2016, creating 18 databases. The models created were evaluated using accuracy, precision and recall rates. Most models obtained accuracy values greater than 80% and precision and recall greater than 0.80, with the best model reaching 96.26% and precision and recall greater than 0.962. The use of the Echo State Network algorithm in the construction of these models, due to its fast training time, emerges as a promising option for predicting cases of arboviruses. This study, in addition to providing an understanding of the areas with the highest incidence and the classification of neighborhoods according to cases, represents crucial steps to mitigate the impact of these diseases on public health.