Enhancing body fat prediction with WGAN-GP data augmentation and XGBoost algorithm.

Area of Science:

• Biomedical Informatics
• Machine Learning
• Data Science

Background:

• Machine learning models can estimate body fat percentage using anthropometric data.
• Limited biomedical data often causes overfitting and reduces predictive accuracy.
• Generative data augmentation is a potential solution for data scarcity.

Purpose of the Study:

• To develop and evaluate a generative data augmentation framework for body fat prediction.
• To enhance the accuracy of machine learning models using limited anthropometric data.

Main Methods:

• Compared Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP), random noise injection, and mixup for data augmentation fidelity.
• Trained and validated XGBoost, Support Vector Regression, and Multi-layer Perceptron models with and without augmentation.
• Assessed model generalization using R², Mean Absolute Error, and Root Mean Squared Error on an independent test set.

Main Results:

• WGAN-GP generated the highest fidelity synthetic data among tested methods.
• XGBoost model performance improved from R² of 0.67 (baseline) to 0.77 with WGAN-GP augmentation.
• Abdominal circumference was identified as the most significant predictor of body fat percentage.

Conclusions:

• WGAN-GP effectively generates realistic synthetic anthropometric data.
• Integrating WGAN-GP synthetic data enhances machine learning model generalization and predictive accuracy.
• This approach provides a robust solution for developing accurate predictive health models in data-limited settings.