Expected Goals Modeling in Football: A Comparative Approach Using Classical and Machine Learning Methods

This paper explores the development of an Expected Goals (xG) model for football using
an open-source dataset from top European leagues. I compare logistic regression with more
advanced techniques, including linear discriminant analysis, bagging, random forest, and neural
networks. The objective is to evaluate model accuracy, calibration, and interpretability while
maintaining transparency and reproducibility. Results show that ensemble methods and neural
networks outperform classical approaches, especially when spatial and contextual features are
properly engineered.