Supervised machine learning algorithms for the classification of obesity levels using anthropometric indices derived...

Area of Science:

• Biomedical Engineering
• Data Science
• Public Health

Background:

• Accurate obesity classification is crucial for public health and clinical decisions.
• Traditional measures like Body Mass Index (BMI) have limitations in distinguishing fat from lean mass.
• Bioelectrical Impedance Analysis (BIA) offers detailed body composition metrics.

Purpose of the Study:

• To evaluate and compare supervised machine learning algorithms for obesity classification.
• To utilize anthropometric indices from BIA for improved obesity assessment.
• To identify key features influencing obesity classification.

Main Methods:

• A cross-sectional study of 5372 adults using BIA-derived anthropometric data.
• Trained and evaluated six machine learning models: Random Forest, Gradient Boosting, KNN, Logistic Regression, SVM, and Decision Tree.
• Assessed model performance using accuracy, F1-score, AUC-ROC, and SHapley Additive exPlanations (SHAP).

Main Results:

• Random Forest achieved the highest accuracy (84.2%), F1-score (83.7%), and AUC-ROC (0.947).
• Fat Mass Index (FMI), Fat-Free Mass Index (FFMI), and BMI were the most influential predictors.
• Sex showed minimal impact on predictive performance.

Conclusions:

• Supervised machine learning, especially Random Forest, demonstrates high accuracy and interpretability in classifying obesity.
• BIA-derived anthropometric data enhances obesity classification beyond traditional BMI.
• These models hold significant potential for improving obesity screening in clinical and community settings.