Results of optimization of random forest model.

Substitution strategy is a fundamental component of tactical adjustments in soccer matches. Since the official introduction of the five-substitution rule in 2020, substitute players have faced increased performance demands. This study examines 2,125 substitution events (excluding goalkeepers) from the 2023 Chinese Football Association Super League (CSL) season, utilizing non-parametric tests (Kruskal-Wallis and Mann-Whitney) to compare soccer player performance differences under different substitution effectiveness conditions. A random forest model is also developed to assess the influence of contextual and player performance variables on substitution effectiveness, aiming to identify optimal decision-making pathways. Results show that, regardless of effectiveness, substitute players display significantly greater high-intensity running distances and passing accuracy than those they replaced. Effective substitutes also demonstrate superior ball recovery. Ineffective substitutions tend to occur earlier in the match, whereas effective and neutral substitutions are more concentrated in later phases. Effective substitutes perform better in medium-to-high-intensity running, passing accuracy, and shooting. Contextual variables contribute more to predicting substitution effectiveness, and the decision pathways suggest prioritizing players with high passing accuracy when leading, while targeted adjustments are needed when trailing.