M.S. Applied Data Science - Capstone Chronicles 2025

5

additive explanations (SHAP) contributions in the xG model. After creating this refined SPPS, we validated its effectiveness by integrating match outcome data (i.e., wins/losses) and implementing logistic regression models to confirm that our weighted metric correlated with winning scenarios. Finally, we conducted comprehensive model comparison using the adjusted SPPS as the target variable, evaluating XGBoost, random forest, gradient boosting, ensemble methods (i.e., stacking and voting classifiers), and neural network architectures based on prediction accuracy Diagram 1 Soccer Tactical Optimization

and cross-validation performance to select the optimal model for player performance prediction. The final implementation provides position-specific performance predictions on a weekly basis throughout the season.

52