Comparing SHAP in RandomForest vs. XGBoost – Which Model Wins?

The Need for Model Interpretability

Both RandomForest and XGBoost provided valuable insights into player valuations, but how do they compare in terms of explainability?

SHAP Summary for XGBoost

• The XGBoost model also highlighted avg_attemptsConcededIbox_Per90_per_game as a key driver of market value, but its SHAP distribution showed even stronger negative impacts for high numbers.

• avg_accurateFwdZonePass_value_per_game had an even greater positive impact in XGBoost compared to RandomForest, suggesting that precise forward passing is crucial in the model's decision-making.

Which Model is Better?

XGBoost often provides more accurate predictions but is harder to interpret due to its complexity. RandomForest is easier to explain but may lack the same predictive power. Depending on the club’s needs—explainability vs. accuracy—the choice of model can differ.

⚡ Final Takeaway: Both models provide useful insights, but teams must balance predictive accuracy with interpretability when applying ML to player valuations.