Metodologia de monitoramento de epidemias: uma abordagem baseada em redes neurais artificiais

Abstract

Dengue fever is a viral infectious disease that is present in more than 100 countries worldwide. In underdeveloped countries such as Brazil, this pathology presents dramatic contours when prevailing socioeconomic factors are added, such as the precarious basic sanitation conditions characteristic of large cities. When we associate this scenario with the Amazon we perceive that the geographic location and climatic conditions of this space contribute to the occurrence of this disease is dimensioned. The Ministry of Health provided data from a survey that found that of the 409,073 reported cases in the North, 106,433 occurred in the state of Pará, where the municipalities with the highest reports of dengue cases are: Belém, Parauapebas, Altamira and Santarém. This work proposes a methodology to monitor epidemics based on the use of Artificial Neural Networks, based on a case study of prediction of dengue cases in the state of Pará. To this end, a system was developed that uses a public database of cases of the disease, of weekly occurrence of the municipalities already mentioned. In addition, it performs the statistical analysis of the series of municipalities showing complexity, and justifying the use of neural networks for this type of problem. It performs the layer adjustments, time window of the trained neural model which in this case is a variation known as recurrent neural network. It implements a module for issuing alerts to detect a sudden increase in new cases of the disease, contributing to the decision-making of public health agencies and their respective actions to control epidemics in the municipalities under study. From our analysis we can conclude that the methodology described in the research is valid for predicting dengue cases using neural networks, anticipating combat actions and contributing to decision making, which can be used by public health managers . And that the use of recurrent neural networks can adjust to the complexity of the series studied. The results demonstrated that the RNA model, for the current scenario, performed well in the epidemiological prediction, reaching satisfactory accuracy