Uma nova abordagem para a estruturação de portifólio para o problema de previsão de casos de dengue

Abstract

Considering the panorama of dengue in the state of Paraná, this study proposes the analysis and organization of a portfolio of machine learning models aimed at forecasting disease cases. The main motivation was the lack of clarity in the parameterization of models in previous studies, compromising the interpretation and reliability of the results. To address this, different machine learning algorithms were explored in various scenarios, including city clustering and parameter tuning, evaluating metrics such as RMSE, MAE, and 𝑅 2 . The results demonstrated the effectiveness of methods such as LSTM and Random Forest, especially in specific contexts. Ultimately, the work contributes with an optimized portfolio, useful for predictions in cities across Paraná, facilitating the practical application of the models and improving decision-making in dengue control.