Leveraging Classification Models to Understand Football Players' and Teams' Characteristics

Abstract

The Expected Goals (xG) is a performance metric used to evaluate a football team's or a player's performance. Simply put, it represents the probability of a scoring opportunity that may result in a goal. This metric suits the low-scoring nature of sports such as football. The score of a match involves randomness and inexplicable factors that skew the data represented by standard metrics and often may not represent the actual performance of an individual or a team; therefore, it would be of more significant benefit to individuals trying to analyse a player or a team to use alternative statistics rather than shots on target, ball possessions percentage, and sprints completed. The xG Model is trained on several key metrics derived from on-field events, corroborating with the historical to measure the probability of a shot being a goal by the common goal. The selection of these features, the size and date of the data, and the model used as the parameters that may affect the model's performance. Using machine learning models to increase the model's predictive performance decreases the vagueness caused by subjective interpretation. This paper proposes an accurate expected goal model trained on a compiled dataset containing data from the FIFA World Cup 2018 and 2022, and the UEFA Champions League 2018-2022, with a total of 768,744 shots taken by the top players take when representing their country and club on the biggest stage. Moreover, this model is explained by using data visualisation tools to obtain an explainable expected goal representation for evaluating a team or player's performance. Moreover, these methods can be generalised to other sports.