Análise espaço-temporal de dados georreferenciados de casos de dengue identificados a partir de rede social no estado do Paraná

Abstract

The rising use of social media by the Brazilian population allows generating an increasing quantity of data. This data allows analyzing many subjects as a study case. This project aims to analyze de data from social media (Twitter) searching to identify contents which indicate dengue cases, by the use of a classifier model developed with sampling and preprocessing tests and analysis based on J48, Support Vector Machine, Maximum Entropy and Multinomial Naive Bayes. A new step used this social media identified cases, using their geographical localization in a DBSCAN density cluster to define the regions with more dengue incidence, forming the social media identified risk regions. To verify the results precision, the results were compared with the Health Ministry official data, making comparisons in different spatialtemporal granularities, to discern where the identification efficacy was. The results allowed to determinate the dengue cases indicative tweets classifier model with 91% of precision. They determinate too the pattern of the efficiency, relating directly with the available data volume, with the spatial-temporal analysis being able to be applied according to the minimum precision desired, varying between 23% and 80% according to the spatial granularity.