B0460
Title: Curve clustering methods and their applications to sports analytics
Authors: Robert Bajons - Vienna University of Economics and Business, Institute for Statistics and Mathematics (Austria) [presenting]
Kurt Hornik - WU Wirtschaftsuniversitaet Wien (Austria)
Abstract: In team sports, such as American football or European football (soccer), players naturally move on the pitch in specific trajectories. Usually, the paths of players on the pitch are determined by specific team tactics, thus interesting analyses can be derived from studying common patterns in these movements. An approach for clustering weighted curves is provided, i.e. curves which may be assigned weights at each observation of the curve, in the context of sports analytics. The weighted K-means approach is simple to implement but relies on substantial preprocessing to be applied to curves. Details of the implementation of the algorithm as well as possible extensions thereof are discussed. Finally, use cases in sports are analyzed, such as an application to pass rush routes in the NFL and the clustering of possession sequences in soccer.