07/03/2022

Folha highlights IMPA model that predicts dengue outbreaks.

Foto: Agência Brasil

Despite being partially overshadowed by the Covid-19 pandemic, dengue fever remains a challenge for Brazil, which recorded a total of 508,200 cases of the disease in 2021. Aware of this reality, a research group at IMPA (Institute for Pure and Applied Mathematics) developed a machine learning model capable of predicting dengue outbreaks in large Brazilian urban centers with 80% accuracy and up to six months in advance. The new system was the subject of a report by journalist Gabriel Alves in Folha de S.Paulo , who highlighted that it "greatly simplifies life and can help various cities prepare for the problem to come."

The tool is available as open source and is the only one fed exclusively with climate data, which facilitates its manipulation for the creation of public policies to combat the disease. The mathematicians spent about two years testing the model with annual climate data from seven capital cities: Aracaju, Recife, São Luís, Belo Horizonte, Salvador, Manaus, and Rio de Janeiro. Through modern computational data processing and machine learning techniques , they managed to improve the accuracy of the model's results from 72% to 80%.

The article, published in the Elsevier journal Expert Systems with Applications , which also made the program code available, was written by IMPA doctoral student Caio Souza in collaboration with the lead researcher of Visgraf (IMPA's Computer Graphics Laboratory), Luiz Velho, IMPA postdoctoral researcher Vitor Rolla, and two former doctoral students from the institute, Lucas Stolerman and Pedro Maia.

A major advantage of the tool is that it requires little data to operate – only temperature and precipitation index – and can be easily applied by municipalities with limited public health funds. “Other, more extensive machine learning approaches require monitoring more data, such as even tweets, and therefore demand more resources from the authorities,” Souza points out.

To use the tool, you don't need in-depth knowledge of the model or advanced knowledge of mathematics or programming. "Anyone with a basic understanding of the Python language can easily run the program," assures Souza. First, you need to populate a spreadsheet with daily temperature and average annual precipitation data, which can be extracted from reliable sources such as the website of the National Institute of Meteorology (INMET) . Then, simply upload this spreadsheet to the code.

The group's objective was to develop a dengue outbreak prediction model that could serve to prevent peaks in cases of the disease in Brazil, and not just remedy them, explains Souza. Therefore, the tool is capable of identifying outbreaks up to six months in advance, and also indicating the best time of year to collect data. The expectation is that the work can help in the development of public health policies for cities, increasing vector control measures during critical periods and improving the prediction of dengue epidemic years.