Predicting differential ratings of perceived exertion (dRPE) in youth soccer using decision tree models.

This study investigated the relationship between external training load metrics obtained from GPS technology and differentiated ratings of perceived exertion (dRPE) in elite youth soccer players, and evaluated the predictive value of decision tree models for estimating exertion across multiple domains. Data were collected over a 17-week in-season period, yielding 537 training observations from 15 players. External load variables included total distance, moderate- and high-speed running, accelerations, decelerations, distances covered during acceleration and deceleration, and impacts, all normalized per minute of training. Internal load was assessed using a modified Borg CR-10 scale with separate ratings for overall exertion, lower-limb effort, breathlessness, and technical-cognitive demand. Decision tree models were developed for each dRPE dimension. Total distance emerged as the strongest predictor of overall exertion, muscular effort, and technical-cognitive strain, while breathlessness was primarily explained by moderate-speed running (14.4-19.8 km·h⁻¹). The models showed satisfactory session-level predictive accuracy within the monitored squad under repeated-session monitoring, with root mean square error values between 0.96 and 1.19 across domains. These results confirm the multidimensional nature of exertion in soccer and indicate the value of combining perceptual measures with interpretable machine learning models to improve individualized monitoring and understanding of training responses in elite youth players.