J Health Popul Nutr. 2026 Jul 12. doi: 10.1186/s41043-026-01401-y. Online ahead of print.
ABSTRACT
BACKGROUND: Childhood stunting remains a major public health concern, reflecting chronic undernutrition and long-term socioeconomic disadvantage. Among school-aged children, stunting is associated with impaired physical growth, reduced cognitive development, and poorer educational outcomes. Traditional statistical approaches, such as logistic regression, have been widely used to examine factors associated with stunting; however, their ability to capture complex and nonlinear relationships is limited. Machine learning (ML) methods provide a flexible alternative for modeling such relationships.
OBJECTIVE: This study aimed to model and classify stunting among school-aged children in Ethiopia using school- and household-level data, and to compare the performance of machine learning algorithms with multivariable logistic regression.
METHODS: A cross-sectional analysis was conducted using secondary data from Round 5 (2016-2017) of the Young Lives study in Ethiopia. Stunting was defined as a binary outcome based on World Health Organization height-for-age Z-score criteria. Several machine learning algorithms, including Random Forest, Support Vector Machine, and Gradient Boosting Machine, were implemented. Model performance was evaluated using accuracy, sensitivity, specificity, F1-score, and area under the receiver operating characteristic curve (AUC). Variable importance measures were used to identify predictors contributing to model performance. To avoid potential circularity, anthropometric variables closely related to the outcome (e.g., child weight) were excluded from the final models.
RESULTS: The Random Forest model demonstrated modestly improved performance compared with logistic regression and other machine learning methods, and its performance was evaluated using accuracy, sensitivity, specificity, AUC, and F1-score. Key predictors included school type, household wealth index, literacy-related indicators, and region of residence. Notable regional variation in stunting classification was observed, suggesting the influence of broader socioeconomic and environmental conditions.
CONCLUSION: Machine learning models, particularly Random Forest, showed slightly better performance than conventional logistic regression in classifying stunting among school-aged children in Ethiopia. The identified predictors highlight the multifactorial and context-dependent nature of stunting. These findings support the use of ML approaches as complementary analytical tools for understanding patterns of child undernutrition, although their application for prediction should be interpreted within the limitations of cross-sectional data.
PMID:42437956 | DOI:10.1186/s41043-026-01401-y
