Prediction of athlete performance based on a gradient regression model.

Area of Science:

• Sports Science and Analytics
• Machine Learning in Sports

Background:

• Conventional statistical models struggle with the complex nonlinearities in athlete physiological, lifestyle, and contextual data.
• Accurate athlete performance prediction is crucial for training design, injury prevention, and talent management.

Purpose of the Study:

• To predict athlete performance scores using tabular data, focusing on predictive validity and generalization.
• To enhance interpretability through SHAP-based explanations and ensure computational efficiency for practical deployment.
• To provide a data-driven platform for identifying key performance determinants and informing sports coaching decisions.

Main Methods:

• Utilized the Kaggle Athlete Performance Prediction Dataset, encompassing demographic, training, physiological, and lifestyle features.
• Applied data preprocessing techniques including imputation, normalization, encoding, and feature engineering.
• Trained a Gradient Regression Model with tenfold cross-validation, comparing its performance against Linear Regression, Ridge Regression, Support Vector Regression, Random Forest, and Neural Networks.

Main Results:

• The Gradient Regression Model achieved a superior R² of 0.923, outperforming Neural Networks (R²=0.901) and Random Forest (R²=0.887).
• Residual and error analysis confirmed minimal bias and variance, while learning dynamics demonstrated efficient convergence and stability.
• The model proved more accurate and interpretable than baseline models.

Conclusions:

• The Gradient Regression Model offers a robust and interpretable solution for athlete performance prediction.
• Findings support applications in individualized training programs and continuous performance monitoring.
• Future research should explore larger, longitudinal studies and hybrid frameworks incorporating biomechanical and psychological data.