Machine Learning-Based Estimation of Sports Performance in Elite Athletes

Achievements and endurance represent distinct aspects of sports performance, influenced by different underlying factors. However, these determinants are not well characterized. To address this gap between achievements and endurance, we evaluated machine learning models to estimate the competition results-based Achievement Score and the absolute cardiorespiratory capacity-based Endurance Score in elite athletes. We analyzed these models using SHapley Additive exPlanations to identify the most influential determinants of each performance component. Altogether, 688 healthy, asymptomatic elite athletes (19.7 ± 6.9 years; males: 75%) with 1194 sports cardiology screening exams were included in our machine learning-based analysis. In a hold-out test set of randomly selected 100 athletes (21.5 ± 9.2 years, males 67%), the best-performing models predicted Achievement Score and Endurance Score with the coefficients of determination values of 0.29 (95% bootstrap CI (0.00, 0.48)) and 0.69 (95% bootstrap CI (0.56, 0.79)), respectively. For Achievement Score, the strongest determinants were training years, weekly training hours, and age. In contrast, Endurance Score was primarily driven by skeletal muscle mass, weight, and peak heart rate. These findings suggest that achievements are predominantly related to training history and experience, whereas endurance is more closely associated with body composition and sports adaptation, partly reflecting its reliance on absolute cardiorespiratory measures.