Bayesian regularized regression of football tracking data through structured factor models

The recent spread of football tracking data motivates the development of statistical tools able to extract and summarize valuable knowledge from the large amount of information available. Factor analysis is routinely used in statistics to reduce dimensionality and when it is applied to a set of regressors it induces regularization that can improve the out-of-sample prediction performances of the linear model. In this article, we propose to use a structured infinite factor model on a set of tracking performance indicators used as covariates of a model for dangerousness of football actions. Such factor model is able to induce a flexible penalty structure on the linear regression model which can be, on the other hand, easily interpreted, providing useful insights in terms of football strategy.