This study examined the effect of contextualised match variables on metabolic power of elite soccer players during English Premier League (EPL) match-play across four seasons (2021/22–2024/25), comparing in-possession (MP_IP) and out-of-possession (MPOP) phases. Match data from 31 male professional soccer players retrieved from 152 regular-season EPL competitive matches was obtained via an optical tracking system and analysed with decision tree regression models. The MP_IP model demonstrated strong predictive accuracy (RMSE = 1.54; MPE = 2.04%), identifying playing position as the dominant predictor (88% of total feature importance). Defenders exhibited the lowest MPIP, while forwards in a 3-5-2 team formation showed the highest values (MP_IP = 15.7 W · kg⁻¹). Conversely, the lowest MPIP values (MPIP = 12.1 W · kg⁻¹) were observed when the study team did not employ a 3-5-2 team formation and faced opponents with a ranking difference of less than eight positions. The MPOP model also demonstrated robust predictive performance (RMSE = 1.59; MPE = 1.30%). Playing position was the most influential factor (44% of total feature importance), with midfielders displaying higher MPOP. Forwards had the lowest MPOP in the second half, while the highest values (MPOP = 17.1 W · kg⁻¹) were observed for midfielders when the team was losing in a 3-5-2 formation. These findings confirm that positional role is the principal determinant of metabolic power in EPL match-play, with midfielders particularly exposed to elevated demands out-of-possession. Tactical structure, match status, and match period further modulate metabolic load, highlighting the need to consider context-specific training strategies.