An innovative RPE-based approach using machine learning to analyse starter and substitute training load in soccer.

Current methods to distinguish starters from substitutes are typically based on playing time. Not considering the physical demands and internal load of different positions can lead to ineffective training and recovery protocols. The aim of this study was to examine whether a k-means clustering approach applied to session-RPE can generate role-specific thresholds that meaningfully differentiate match load profiles between starters and substitutes. We analysed 1,450 player-matches from four professional Italian Serie A teams, using video match analysis to measure total distance (TD) and high-intensity activities: metabolic power events (MPE), high-speed running (HSR), and sprint running (SR). Players were divided based on the role as follows: forwards (FWs), midfielders (MFs), full-backs (FBs), and centre-backs (CBs). Individualized sRPE zones (low, medium, high) were identified with the K-means clustering approach discriminating starters from substituted. FWs, MFs, and FBs were considered substituted, and compensatory training was recommended when the sRPE was within the medium sRPE zone or lower (FWs ≤ 695 a.u., MFs ≤ 711 a.u., and FBs ≤ 726 a.u.). Compensatory training particularly focused on SR was recommended at sRPE = low for FWs (≤ 326.1 a.u.), and at a sRPE ≤ medium for MFs (≤ 711 a.u.). CBs were defined as starters when reporting sRPE values ≥ medium sRPE (> 446 a.u.), and SR training was always recommended. The proposed sRPE-based k-means approach distinguishes fatigued from non-fatigued players, guiding decisions about who should prioritise recovery. Role-specific sprint recommendations help coaches provide appropriate high-velocity exposure to prevent hamstring injuries.