Modelling Football Players’ Field Position via Mixtures of Gaussians with Flexible Weights

This contribution provides an empirical analysis on player position on the field throughout a football match. To identify potential phases of the game, a Bayesian mixture of experts model is defined where the relationship between the location of a player and their teammates is specified using a mixture via a flexible formulation of the component weights, which are modelled as smooth functions of concomitant covariates. The analysis shows that player positions can be decomposed into phases of play that are related to the other teammates’ locations on the field.