Data-augmented machine learning for personalized carbohydrate-protein supplement recommendation for endurance.

Area of Science:

• Sports Science
• Machine Learning
• Nutritional Biochemistry

Background:

• Carbohydrate-protein supplementation is known to improve endurance performance.
• Individual responses to supplementation vary significantly due to unique personal characteristics.

Purpose of the Study:

• To develop a predictive machine learning framework for personalized carbohydrate-protein supplementation strategies.
• To address data scarcity in predicting athletic performance using Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP).

Main Methods:

• Utilized 231 rowing trials with 46 input features (baseline characteristics, dietary intake).
• Employed a hybrid feature selection method and WGAN-GP for data augmentation.
• Trained regression models (XGBoost, SVR, MLP) to predict rowing performance.

Main Results:

• Identified 21 key performance indicators from 46 initial inputs.
• The XGBoost model, augmented with WGAN-GP data, achieved high predictive accuracy (R² = 0.53).
• Key predictors included body weight, explosive power, and nutritional inputs.

Conclusions:

• A data-augmented machine learning approach effectively models individual responses to supplementation.
• The developed framework offers a data-driven pathway for personalized nutritional strategies to optimize athletic performance.