SHAP-enhanced machine learning identifies modifiable obesity predictors across adolescent weight groups: A 2021 YRBSS...

Area of Science:

• Public Health
• Data Science
• Adolescent Medicine

Background:

• Adolescent obesity is a global public health crisis.
• Traditional methods often fail to capture the complexity of obesity causes.
• This study employs advanced machine learning to identify key drivers of adolescent obesity.

Purpose of the Study:

• To identify and rank the primary causes of adolescent obesity using machine learning.
• To provide data-driven insights for developing effective obesity prevention strategies.
• To analyze the interplay of behavioral, dietary, sleep, and substance use factors in adolescent obesity.

Main Methods:

• Utilized data from the 2021 Youth Risk Behavior Surveillance System (YRBSS) for adolescents aged 12-18.
• Applied Random Forest and XGBoost machine learning models to analyze risk factors.
• Employed SHapley Additive exPlanations (SHAP) for enhanced model interpretability.

Main Results:

• Key predictors identified: breakfast frequency, moderate-to-vigorous physical activity (MVPA) days, sleep duration, fruit intake, screen time, and vaping.
• Machine learning models showed modest discriminative ability (AUC ~0.58).
• Lower breakfast frequency, fewer MVPA days, shorter sleep, lower fruit intake, and increased screen time were linked to higher obesity risk.

Conclusions:

• Machine learning models successfully identified critical behavioral and lifestyle predictors of adolescent obesity.
• Public health interventions should emphasize promoting daily breakfast, fruit consumption, regular physical activity, adequate sleep, reduced screen time, and vaping prevention.
• Addressing these factors is crucial for mitigating the rising rates of adolescent obesity.