Análise de séries temporais e utilização de algoritmos de machine learning para a predição de casos de dengue em Santa Maria (RS)

Abstract

The global incidence of dengue has increased dramatically in recent decades, according to the World Health Organization. In the Americas, more than 500 million people are at risk of contracting the disease. With local variations of risk influenced by precipitation, temperature, rapid unplanned urbanization and other constraints, countries such as Brazil have greater susceptibility to the virus. In this sense, considering the influence of environmental determinants and climatic factors in endemics, the importance of using machine learning algorithms was identified in order to describe the behavior and dynamics of dengue transmission in Santa Maria (RS) to generate forecasts about the number of cases in the city. The work is based on the use of time series and the methodology consists of implementing an integrated autoregressive model of moving averages (ARIMA) and the Categorical Boosting algorithm to estimate the number of cases that will occur in an interval of 1 to 3 months, being presented comparisons between the predictions of the two algorithms. The models receive as input historical data series of dengue and the record of climatic variables over time to produce the predictions. For a 12-week period ahead, tests show better performance for CatBoost.