Aplicação de machine learning na análise de factores associados as uniões prematuras em Moçambique

Abstract

Early unions constitute a phenomenon of social and public-health relevance in Mozambique, with negative implications for the educational and economic development and the overall well- being of girls. This research aimed to apply Machine Learning techniques to the analysis of the factors associated with this phenomenon, complementing them with classical statistical methods. Data from the Mozambique Demographic and Health Survey (DHS, 2022–23) were investigated, with a descriptive characterization and a bivariate analysis conducted to examine associations between the explanatory variables and the occurrence of early unions. A Logistic Regression model was then fitted, accounting for the complex survey design, and the Extreme Gradient Boosting (XGBoost) algorithm was applied to estimate the probability of occurrence and to compare the predictive performance of the approaches. The XGBoost model was optimi- zed through hyperparameter tuning using grid search and cross-validation. Logistic Regression achieved an area under the ROC curve (AUC) of 0.82, showing good discriminative ability, while XGBoost achieved a mean AUC of 0.832 in cross-validation and 0.817 on the test set, indicating similar overall performance, with a slight advantage during validation. The variable- importance analysis identified age at first sexual intercourse, marital status, level of education, and socioeconomic factors as the main determinants, suggesting that the phenomenon results from the interaction of demographic, social, and behavioral factors. It is concluded that the ap- plication of Machine Learning techniques, combined with classical statistical methods, proved adequate for analyzing the factors associated with early unions and assessing the predictive per- formance of the models, providing relevant empirical evidence to support the design of public policies and intervention strategies aimed at preventing this phenomenon in Mozambique.