Explainable AI in Football: Enhancing XGBoost interpretation with SHAP, Counterfactuals and LLM

Football is a complex game, and recently, data-driven decisions have become increasingly popular for developing strategies and understanding player gameplay among coaches and club analysts. Numerous studies have utilized machine learning techniques to analyze football data and derive insights into player behavior. However, this implementation presents the data and results in a numerical output. To understand it and take the advantage of this result it is still a tricky part for the coaches, as they are not in the field of the data scientist and still wonder how the model makes the decision and also the numerical data is not self-explainable for them, hence the final result still cannot be entirely utilized as it remains unclearly understood to the users. The project aims to address this gap and make the output as understandable as possible using new tools, such as Large Language Model(LLM), explainable Artificial Intelligence, and machine learning models. Extreme Gradient Boosting (XGBoost) is chosen as the decision made by this model is completely executed in the black box, and to explore it and make it explainable using tools such as Partial Dependence Plot (PDP), counterfactual explanations, and the Shapley Additive exPlantions (SHAP) library. Furthermore, To have all these tools available to the user at one stop, a web-based application with Large Language Model(LLM) text is easy for coaches to understand and interpret.