Interpretable machine learning for identifying adolescent obesity risk and identifying key determinants

Area of Science:

• Public Health
• Computational Biology
• Pediatrics

Background:

• Adolescent obesity is a growing public health concern with complex contributing factors.
• Understanding individual, family, and school influences is crucial for effective prevention strategies.
• Machine learning offers advanced tools for analyzing large datasets to identify key risk factors.

Purpose of the Study:

• To apply interpretable machine learning (ML) to identify and prioritize factors associated with adolescent obesity.
• To establish specific risk thresholds for targeted interventions.
• To analyze data across individual, family, and school domains.

Main Methods:

• Utilized data from the China Education Panel Survey (CEPS) involving 7,397 adolescents.
• Developed and evaluated six ML models: Support Vector Machine (SVM), XGBoost, LightGBM, Logistic Regression (LR), Random Forest (RF), and Multilayer Perceptron (MLP).
• Employed SHapley Additive exPlanations (SHAP) analysis for interpreting the best-performing model and assessing feature contributions.

Main Results:

• The LightGBM model achieved the highest accuracy (0.8788) in classifying adolescent obesity.
• Key predictors identified include sedentary time, school ranking, academic workload, birth weight, body image, family economic status, school location, and household registration.
• Sedentary behavior was the most significant predictor, with risk thresholds identified such as >5 hours of weekend sedentary time and birth weight >4.0 kg.

Conclusions:

• Interpretable ML effectively identifies critical predictors of adolescent obesity.
• Interventions should prioritize reducing sedentary behavior, moderating academic workload, and enhancing body image perception.
• Family and school environments are vital components in adolescent obesity prevention efforts.