EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
2550. Some novelty on the xG model for Football Analytics
Invited abstract in session TC-16: Football analytics, stream OR in Sports.
Tuesday, 12:30-14:00Room: 19 (building: 116)
Authors (first author is the speaker)
1. | rodolfo metulini
|
Department of Economics, University of Bergamo | |
2. | Mattia Cefis
|
Department of Economics and Management, University of Brescia |
Abstract
Within the realm of football analytics, an objective is to enhance the predictive capabilities and exploit the emerging tool known as the expected goal (xG) model for pratical needs. The xG model aims to classify each shot, predict the final score of a game and evaluate strikers and goalkeepers performance. In previous works we proposed to match event data and composite indicators of player performance derived through Partial Least Squares - Structural Equation Model (PLS-SEM).
The aim of this work is twofold: leveraging a dataset comprising tracking data from the Italian “Serie A” related to 2022/2023 season, we introduce to the existing model some original features, such as the Kos angle. Our results showcase intriguing findings in comparison to a benchmark (Understat), particularly in certain metric indices. Furthermore, the significance of performance composite indicators from PLS-SEM and specific tracking variables for the xG model is established.
Afterwards, we exploit the performance of the model proposing a combination of machine learning and game theoretical approaches to assess the marginal contribution of single players in determining the expected goal and the final game score. In particular, to achieve this latter goal, we adopt our xG as the cohesion function of a (generalized) Shapley value. By doing so, an instrument to jointly estimate the probability to score the goal and evaluate the role of each player in determining such probability is proposed.
Keywords
- OR in Sports
- Machine Learning
Status: accepted
Back to the list of papers