In the field of football analytics, the goal is to improve (in terms of prediction performance) one of the emerging tools: the expected goal (xG) model. With this final aim, data from different sources have been merged: tracking data, match event data and some players’ performance composite indicators obtained using a Partial Least Squares - Structural Equation Model (PLS-SEM) approach. Using a sample of match data relying to season 2019/2020 of the Italian Serie A, composed by 1 outcome variable (i.e. the GOAL) and 22 features, a logistic regression model was applied on different scenarios for sample balanced techniques. Results seem to be interesting in terms of sensitivity, F1 and AUC metrics, compared with a benchmark. In addition, some original performance composites and tracking variables introduced are significant for the classification model.
A PLS-SEM Approach for Composite Indicators: An Original Application on the Expected Goal Model
Mattia Cefis
2024-01-01
Abstract
In the field of football analytics, the goal is to improve (in terms of prediction performance) one of the emerging tools: the expected goal (xG) model. With this final aim, data from different sources have been merged: tracking data, match event data and some players’ performance composite indicators obtained using a Partial Least Squares - Structural Equation Model (PLS-SEM) approach. Using a sample of match data relying to season 2019/2020 of the Italian Serie A, composed by 1 outcome variable (i.e. the GOAL) and 22 features, a logistic regression model was applied on different scenarios for sample balanced techniques. Results seem to be interesting in terms of sensitivity, F1 and AUC metrics, compared with a benchmark. In addition, some original performance composites and tracking variables introduced are significant for the classification model.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
