Development of Machine-Learning-Based and Deep-Learning-Based Models for Predicting Korean Adolescents’ Overweight and Obesity

Objectives: Overweight and obesity in adolescents are a leading health concern worldwide, including in South Korea. This study aims to develop predictive models for overweight and obesity in Korean adolescents using 11 machine learning and deep learning techniques: logistic regression, ridge, LASSO, elastic net, decision tree, bagging, random forest, AdaBoost, XGBoost, support vector machine, and fully connected layer models. Methods: We used 43,268 records (Grades 7–12) from the 16th Korean Youth Risk Behavior Web-Based Survey. The survey data included 71 factors that may influence overweight and obesity among adolescents, encompassing sociodemographic characteristics; dietary habits; physical and psychological health; behavioral problems; and family, peer, and school factors. Results: The machine learning and deep learning algorithms displayed significantly superior performance in predicting overweight and obesity among Korean adolescents when compared to logistic regression. XGBoost was particularly effective: accuracy 0.8403, recall 0.6351, precision 0.6497, F1 0.6423, and area under the curve 0.8982. Conclusion: The machine learning and deep learning models developed in this study to predict overweight and obesity in Korean adolescents hold potential for use in practical applications in social work settings.