Understanding the relationship between fitness, anthropometric characteristics, and internal load is essential for optimising training and reducing injury risk in youth athletes. This study classified senior youth badminton players into exertion groups using decision tree (DT) models, guided by rate of perceived exertion (RPE). Seventy-three players (mean experience = 6 years) completed physical assessments, and RPE was recorded during match play. K-means clustering identified two exertion groups: Low Exertion Group (LEG, n = 36) and Moderate Exertion Group (MEG, n = 37). A DT regression model predicted RPE with high accuracy (R2 = 0.87 training, 0.95 test; MAE = 0.25 training, 0.07 test). Push-up performance was the most significant split variable (>25 reps for MEG), with additional predictors including standing broad jump, single-leg wall-sit, vertical jump, and stork balance test. SHAP analysis highlighted stork balance, single-leg wall sit, push-ups, height and hand grip strength as key contributors across groups. For LEG, static balance and muscular endurance emerged as dominant positive predictors. These findings accentuate the importance of neuromuscular coordination and strength in modulating exertion. Interpretable machine learning models offer practical tools for developing individualised, data-informed training strategies that enhance performance while accounting for athletes' unique physical profiles.
Keywords: Youth badminton players; anthropometric characteristics; decision tree model; fitness parameters; machine learning; workload management.