Quantifying Individual Performance in Football : An XGBoost Approach using Match-Level Event Data

Sammanfattning

The aim of this thesis is to investigate if XGBoost can be used to create a meaningful player rating system for football players based on aggregated event data, using the match result as a proxy for player performance. The model is trained to estimate the probability of a player winning a specific match, with its summarized statistics from said match acting as features in the model. To evaluate the performance of the XGBoost model, it is compared to a simple logistic regression model. The results show that the XGBoost model slightly outperforms the logistic model, although the difference in predictive ability is small. The study also shows that offensive metrics, closely correlated with the probability of scoring, have the highest variable importance in the XGBoost model. The thesis finally concludes that, although its predictive capacity is limited, the XGBoost model is able to generate reliable player match ratings and can be implemented to act as a measurement of the relative performance of football players.