Comparative Analysis of Predictive Models for Childhood Obesity

A recent study compared various machine learning and statistical approaches to identify the determinants of overweight and obesity among children and adolescents in the United States. The research analyzed data from the 2021 National Survey of Children's Health, involving 18,792 participants aged 10 to 17 years.

Methodologies Compared

Several models were evaluated, including logistic regression, random forest, gradient boosting (XGBoost and LightGBM), multilayer perceptron (MLP), and TabNet. Predictors included diet, physical activity, sleep, parental stress, socioeconomic conditions, adverse experiences, and neighborhood characteristics. Performance metrics used were AUC, accuracy, precision, recall, F1 score, and Brier score.

Results and Implications

The results showed that the discrimination ability of the models ranged from 0.66 to 0.79. Logistic regression, gradient boosting, and MLP demonstrated the best balance between discrimination and calibration. Boosting and deep learning slightly improved recall and F1 score, but no model proved uniformly superior. Performance disparities persist across racial and socioeconomic groups, suggesting the need to improve data quality and equity-focused surveillance, rather than increasing algorithmic complexity.