Análise dos perfis de risco do trabalho infantil no Brasil com técnicas de machine learning

Abstract

This study examines the risk profiles associated with the incidence of child labor in Brazil, aiming to identify the socioeconomic, demographic, and household factors linked to the likelihood of early entry into the labor market. Using a sample of children aged 5 to 13 years, the analysis relies on microdata from the 2024 edition of the Continuous National Household Sample Survey (PNADC). The methodological approach applies the Extreme Gradient Boosting (XGBoost) supervised learning algorithm, combined with interpretability techniques based on SHAP (SHapley Additive exPlanations) values, which make it possible to assess the individual contribution of each explanatory variable to the model’s predictions. The results indicate that the child’s age and sex, household location, per capita income, and the educational level and occupational sector of the household head—particularly when associated with agricultural work—play a decisive role in shaping risk profiles. Older children, boys, those living in rural areas, and those in households with lower income and human capital exhibit higher predicted probabilities of child labor. The findings demonstrate the potential of interpretable machine learning tools to uncover complex patterns, offering valuable support for the design of policies aimed at preventing and reducing child labor in the country.