Sports are characterized by unique rules, environments, and tasks, but also share fundamental similarities with each other sport. Such between-sports parallels can be vital for optimizing talent transfer processes. This study aimed to explore similarities between sports to provide an objective basis for clustering sports into families by means of machine learning. An online survey was conducted, garnering responses from 1,247 coaches across 36 countries and 34 sports. The survey gauged the importance (0 = not important 10 = important) of 18 characteristics related to the sport and the athlete performing in that sport. These traits formed the basis for the categorization of a sport by means of machine learning, particularly unsupervised clustering, and the LIME feature explainer. Analysis grouped 34 sports into five clusters based on shared features. A similarity matrix illustrated the degree of overlap among sports. The application of unsupervised clustering emphasized the lack of a single overarching attribute across sports, marking a shift away from traditional clustering approaches that rely on a limited set of characteristics for talent transfer. The results highlight the importance of identifying common sports for talent transfer, which could prove advantageous in guiding athletes towards new sporting directions.
Keywords: clustering; coaches; experiential knowledge; machine learning (ML); questionnaire; sports; talent transfer.
© 2024 Teunissen, De Bock, Schasfoort, Slembrouck, Verstockt, Lenoir and Pion.