Answers are grounded in this paper and cite their sources.
Effective epidemic management requires proactive measures, making the accurate prediction of weekly dengue cases in specific regions essential for prevention and control strategies. In this study, we evaluated the effectiveness of classical statistical techniques and machine learning methods in forecasting the number of weekly dengue cases across 27 Brazilian federative units. For each unit, we explored a multivariate one-step prediction strategy, applying wavelet filtering to the features to enhance signal decomposition and improve predictive performance. We investigated the univariate LightGBM model trained on data from 26 cities (cross-learning) and validated it individually in each federative unit using the univariate leave-one-out (LOO) technique with one-step predictions. Additionally, we demonstrated the model’s generalizability by training LightGBM on data from all Brazilian federative units and validating it in a different geographic location, San Juan. The LightGBM LOO model exhibited superior generalization compared to other shallow, deep, and foundation models, including TimeGPT-1 and MOIRAI.
No takes yet. Share an insight, caveat, or question.
Zanardo et al. (2025) studied this question.