Van Oirbeek, RobinMwana-Nteba, LoïcLoïcMwana-Nteba2025-07-082025-07-082025-05-2720252025-05-28https://hdl.handle.net/2078.2/43550This work addresses predictive modeling of football analytics by introducing a transparent and reproducible methodology to label and predict meaningful offensive and defensive actions that lead to high-quality shots. A rule-based labeling function was created based on the 2017/2018 Premier League season's Wyscout event data to identify offensive and defense-related meaningful actions. These labels were then used to develop player-level statistics and a composite player performance score, which were found to have moderate correlations with extrinsic measures like player wages and Ballon d’Or rankings. In a second phase, a number of machine learning models such as LightGBM, XGBoost, and Random Forest were trained to make predictions from these contribution labels. The analysis compared varying preprocessing approaches, tested class imbalance solutions, and applied SHAP values to facilitate interpretability. The results show the power of the combination of interpretable labels and structured models to evaluate player performance. The paper contributes to the possibility of transparent model pipelines in sports analytics and sets the stage for future expansions through richer contextual information.text::thesis::master thesis