How Partial Dependence Plots (PDPs) Reveal XGBoost’s Decision-Making

Understanding PDPs in XGBoost

Partial Dependence Plots (PDPs) allow us to isolate how individual features impact player valuation, providing a more detailed look at how the XGBoost model makes its predictions.

Key PDP Insights from XGBoost

📌 Big Chances Created Drive Value

• The PDP for avg_bigChanceCreated_per_game shows a clear upward trend: players who create more big chances see a sharp increase in their predicted market value.

• The steepest rise occurs when moving from 0.5 to 1.5 big chances per game, suggesting that elite playmakers significantly increase their worth with consistent contributions.

📌 The Impact of Open Play Contributions

• The relationship between avg_attOpenPlay_value_per_game and market value is nonlinear: players who contribute to open play attacks see gains up to a certain point, after which the effect plateaus.

• This suggests that there is an optimal level of attacking involvement beyond which additional contributions do not significantly raise a player’s valuation.

📌 Team Success Has a Threshold Effect

• Team_points shows an interesting trend: a rise in points has a strong positive effect up to a certain level (~70 points), beyond which additional points have diminishing returns.

• This aligns with market behavior—players in mid-to-high-performing teams benefit the most in terms of value appreciation.

How Teams Can Leverage PDP Insights

🔍 PDPs offer clubs a predictive tool for scouting. By identifying tipping points where certain metrics maximize valuation, clubs can refine player development programs and optimize transfer strategies.

🔜 Next: SHAP Values – A Player-by-Player Breakdown of Market Value Drivers!