Previsão de nascidos vivos nas regiões de saúde do Brasil através de modelos de aprendizado de máquina baseados em árvore

Abstract

Forecasting tree-based models are a type of predictive modeling technique that uses decision trees to make predictions about future values or events. These models are good choices due to their ability to model non-linear relationships, which is why they were applied to predicting live births with multiple covariates. The study uses data from the Brazilian Ministry of Health to train and evaluate forecasting models, following the guidelines of the Ministry’s expectations and needs for public policy planning. The study uses data from all 450 microregions in Brazil with records between the years 2000 and 2020. The objective is to train a tree-based model with all months between 2000 and 2018 to evaluate the performance of predicting the number of births over of the years 2019 and 2020. LightGBM, XGBoost and Catboost were evaluated and compared with AutoARIMA and simple linear regression. LightGBM performed slightly better than other evaluated models, achieving a MAPE of 0.0797, with a more consistent performance over the 24-month forecast horizon. The results show that tree-based models are reliable for handling multiple covariates and can be a useful tool for public policy planning. Keywords <Time Series Forecasting, Long-Horizon Forecasting, Decision Trees, Maternal Mortality