The principles of tactical formation identification in association football (soccer) — a survey

This paper reviews the principles employed to identify team tactical formations in association football, covering over two decades of research based on event and tracking data. It first defines formations and discusses their history and importance. It then introduces the preprocessing and team/position-level principles. Preprocessing includes match segments and normalized locations followed by data representation using various options, such as average locations, hand-engineered features, and graphs for the team-level and relative locations, distributions, and images for the position-level approaches. Either of them is later followed by applying templates or clustering. Among the limitations for future research to address is the reliance on spatial rather than temporal aggregation, which bases formation identification on newly introduced coordinates that may not be available in raw tracking data. Assuming a fixed number of outfield players (e.g., 10) fails to address scenarios with fewer players due to red cards or injuries. Additionally, accounting for phases of play is crucial to provide more practical context and reduce noise by excluding irrelevant segments, such as set pieces. The existing formation templates do not support arrangments with more or fewer players in each horizontal line (e.g., 6-3-1). On the other hand, clustering forces new observations to be described with previously learned clusters, preventing the possibility of discovering emerging formations. Lastly, alternative evaluation methods should have been explored more rigorously, in the absence of ground truth labels. Overall, this study identifies assumptions, consequences, and drawbacks associated with formation identification principles to structure the body of knowledge and establish a foundation for the future.