Explainable Machine Learning Approaches For Predicting Maximal Oxygen Uptake (VO2max) In Professional Male Soccer Players

PURPOSE: Non-exercise-based models for predicting maximal oxygen uptake (VO2max), particularly those employing machine learning (ML), are limited, with no models specifically tailored for professional male football players. METHODS: Data from 800 male professional football players were utilized, with VO2max (mL/kg/min) as the dependent variable. Independent variables encompassed 53 anthropometric, physical, and physiological factors, including but not limited to age, BMI, PlayerLoad, heart rate, lactate levels, and other relevant metrics. The Synthetic Minority Oversampling Technique (SMOTE) addressed class imbalance. Feature selection was performed using Boruta and Least Absolute Shrinkage and Selection Operator (LASSO). Five ML algorithms—Linear Regression (LR), Random Forest (RF), Gaussian Naive Bayes (GNB), Multi-layer Perceptron (MLP), and Support Vector Regression (SVR)—were trained with 10-fold cross-validation. Model interpretation was conducted using Shapley Additive Explanations (SHAP) to assess feature importance. Model performance was evaluated based on R2 and root mean squared error (RMSE). RESULTS: The mean age of participants was 27.4 ± 4.5 years, with a VO2max of 61.5 ± 6.0 mL/kg/min. ML models exhibited RMSE values ranging from 2.8 to 3.5 mL/kg/min. The SVR model outperformed the others, achieving an R2 of 0.85 and an RMSE of 2.9 mL/kg/min. SHAP analysis identified PlayerLoad and lactate as the most influential predictors of VO2max. CONCLUSIONS: ML models, particularly SVR, offer a robust method for predicting VO2max in professional football players. Key predictors such as PlayerLoad and lactate can enhance endurance performance monitoring and individualized training strategies. These models represent a promising tool for non-exercise-based VO2max estimation in professional football players.