Beyond the pitch predictive and explainable AI applications in football analytics

Abstract

The application of machine learning in football analytics has significantly advanced, yet challenges remain in achieving a balance between predictive accuracy and interpretability. This thesis investigates the effectiveness of predictive models and explainable AI (XAI) techniques in forecasting football match outcomes and providing actionable insights for managerial decision-making. Historical match data, player statistics, and ELO ratings from the English Premier League (2017–2024) serve as the foundation for developing and evaluating machine learning models, including Random Forest, Gradient Boosting, and XGBoost. Explainable AI techniques, such as SHAP (SHapley Additive exPlanations), are applied to interpret model outputs both globally and locally, revealing key predictors of match outcomes, including ELO differences, expected goals, and positional metrics. Formation simulations are utilized to assess the impact of various team setups on predicted outcomes, offering practical insights into tactical decision-making. Results indicate that XGBoost achieves the highest predictive accuracy (55.2%), comparable to bookmaker odds provided by Bet365. SHAP visualizations enhance the interpretability of model outputs, identifying the features most influential in determining predictions and supporting more transparent decision-making processes. This research demonstrates the potential of combining predictive analytics with XAI to optimize tactical planning, improve player deployment, and refine strategic operations. By bridging the gap between complex predictive models and actionable insights, the study provides a robust framework for advancing data-driven decision-making in football analytics.

Date of Award	31 Jan 2025
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Ana Marisa Mendes Gonçalves Vinhais Guedes (Supervisor)