Evaluation of xG Models in Football

Data from: Understat

Report published in Pages

In this project, I explored various statistical models to evaluate expected goals (xG) in football, aiming to identify the most suitable model. The evaluation was performed by considering Open Play and Set Piece actions separately, further dividing the data between Foot and Head shots.

Models used:

• Logit (logistic regression)
• Logit with quadratic interactions
• Random Forest
• Bagging
• Neural Network (Neural Net)

Methodology

There is no "perfect model" that works in all situations, and the choice of the best model depends on the context and specific goals. To assist in the evaluation, I have included both ROC Curve and Calibration Plot for each model in the Final Results folder, alongside the xG data calculated by Understat used as a reference.