Expanded Predictors Improve Classification of Adolescent Overweight and Obesity

A cross-sectional analysis of 17,814 U.S. adolescents found that adding broader predictor information improved machine-learning classification of current overweight or obesity. The study used 2023 Youth Risk Behavior Surveillance System data from participants aged 12 to 18 years and estimated that 31.8% had overweight or obesity. Because body mass index and candidate predictors were measured in the same survey cycle, the models classified adolescents’ status at that time rather than predicting future obesity.

Researchers used survey-aware multivariable logistic regression to identify variables independently associated with overweight or obesity, then compared Random Forest and XGBoost under base and extended predictor frameworks. On a held-out test set, the extended Random Forest model increased its area under the receiver operating characteristic curve from 0.6241 to 0.6894. The corresponding XGBoost value rose from 0.6217 to 0.6724. These results indicate that the expanded information improved classification within this dataset and analysis; they do not show that changing an adolescent’s characteristics would cause a change in body mass index status.

Male sex was associated with higher odds of overweight or obesity, as were Black or African American and Hispanic/Latino race or ethnicity. Higher moderate-to-vigorous physical activity and more frequent breakfast intake were associated with lower odds. The Random Forest model with extended predictors, designated RF-Extended, was selected for calibration and interpretability analysis. Isotonic recalibration preserved its discrimination while improving the reliability of its probability estimates. The Brier score decreased from 0.2126 to 0.1960, and expected calibration error fell from 0.1160 to 0.0213.

Exploratory SHAP analysis identified sex, race or ethnicity, physical activity, sleep duration, age and breakfast frequency among the features contributing most strongly to RF-Extended output. This analysis described contributions within the fitted model and did not establish biological mechanisms or causal determinants. The study’s scope remains important: it was based on one survey cycle, and its models estimated current status rather than forecasting later obesity. Prospective and external validation would be required before considering applications in other populations, including Vietnamese adolescents. The findings therefore provide a survey-aware evaluation of expanded predictors, two machine-learning algorithms, probability calibration and feature attribution, rather than evidence that the models are ready for screening or prevention use.

Subject of Research: Survey-aware machine-learning classification of current overweight and obesity among 17,814 U.S. adolescents using 2023 YRBSS data

Article Title: Broader factors associated with current adolescent overweight/obesity status in YRBSS 2023: Survey-aware analysis and calibrated AI-assisted classification

Article References: Thanh, T. H., Thi, N. N., & Nguyen Khac, T. (2026). Broader factors associated with current adolescent overweight/obesity status in YRBSS 2023: Survey-aware analysis and calibrated AI-assisted classification. PLOS One, 21(10), e0360155. https://doi.org/10.1371/journal.pone.0360155

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0360155

Keywords: adolescent overweight, adolescent obesity, machine learning, Random Forest, XGBoost, model calibration, physical activity, breakfast intake, YRBSS, SHAP analysis, Broader, factors