3 69 MACHINE LEARNING SUPPORTING SUBSTITUTIONS IN SOCCER variables. The appliance of machine learning enables the prediction of a player’s physical performance in an early stage in the match whereby the more sensitive type-2 variable outperforms the type-1 variables in the precision of the prediction. 5.1 Practical Implications These findings show that it is possible to identify underperforming players in an early stage in the match. Applying machine learning in combination with monitoring the energy expenditure in power category during the match enables real-time support for the coach to decide on substitutions. For the nature of the game is the same for many leagues, monitoring expenditure in power category can be of use in many other environments than Dutch elite soccer. A precondition for the support system is to set up a dataset per player, which allows for tracking during the season and machine learning. Future research to refine the machine learning models may include the influence of contextual factors such as home-away, score, ranking, and player position. Acknowledgments We like to thank J van Norel, S.V.B. Vitesse, Arnhem, The Netherlands for sharing the competition data.
RkJQdWJsaXNoZXIy MjY0ODMw