AUTHOR=Araujo-Moura Keisyanne , Souza Letícia , de Oliveira Tiago Almeida , Rocha Mateus Silva , De Moraes Augusto César Ferreira , Chiavegatto Filho Alexandre 

TITLE=Prediction of Hypertension in the Pediatric Population Using Machine Learning and Transfer Learning: A Multicentric Analysis of the SAYCARE Study

JOURNAL=International Journal of Public Health

VOLUME=Volume 70 - 2025

YEAR=2025

URL=https://www.ssph-journal.org/journals/international-journal-of-public-health/articles/10.3389/ijph.2025.1607944

DOI=10.3389/ijph.2025.1607944

ISSN=1661-8564

ABSTRACT=Objective: To develop a machine learning (ML) model using transfer learning (TL) to predict hypertension in South American children and adolescents.Methods: Data from cohorts in seven South American cities were analyzed. A CatBoost model trained on children’s data was adapted to adolescents using TL. Performance was evaluated with standard metrics.Results: Among children, 88.9% had normal blood pressure, while 14.1% had elevated blood pressure (EBP). In adolescents, 92.5% had normal blood pressure, and 7.5% had EBP. For children, Random Forest, XGBoost, and LightGBM showed high accuracy (0.90), with XGBoost and LightGBM achieving superior recall (0.50) and AUC-ROC (0.74). Adolescent models without TL performed poorly (AUC-ROC 0.46–0.56). TL improved performance significantly, with CatBoost achieving an AUC-ROC of 0.82, accuracy of 1.0, and recall of 0.18.Conclusion: Dietary predictors of EBP included soft drinks, filled cookies, and chips, with higher intake in adolescents. ML with TL effectively identified hypertension risks, highlighting the importance of early dietary interventions to prevent hypertension and promote cardiovascular health in pediatric populations.