Differentiating movement styles in professional tennis: A machine learning and hierarchical clustering approach.

Purpose: Recent explorations of tennis-specific movements have developed contemporary methods for identifying and classifying changes of direction (COD) during match-play. The aim of this research was to employ these new analysis techniques to objectively explore individual nuance and style factors in the execution of COD movements in professional tennis. Methods: Player tracking data from 62 male and 77 female players at the Australian Open Grand Slam were analysed for COD movements using a model algorithm, with a sample of 150,000 direction changes identified. Hierarchical clustering methods were employed on the time-motion and degree characteristics of these direction changes to identify groups of different COD performers. Results: Five unique clusters, labelled "Cutters", "Gear Changers", "Lateral Changers", "Balanced Changers" and "Passive Changers" were identified in accordance with their varying speed, acceleration, degree and directionality of change features. Conclusions: Player COD clustering challenge previously held assumptions regarding on-court movement style, highlighting the complexity and variation in the sport's locomotion demands. In practice, the speed, acceleration, directionality and degree of change characteristics of each COD style can facilitate athlete profiling and the specificity of training interventions. We used machine learning techniques and cluster analysis methodology to explore the time motion characteristics of direction change skill in professional tennis. We present five unique types of change of direction style in professional tennis players. These include "Cutters", "Gear Changers", "Lateral Changers", "Balanced Changers" & "Passive Changers". These style classifications were established in accordance with their varying speed, acceleration, degree and directionality of change features. We show that the application of machine learning techniques to player tracking data can facilitate a more intricate understanding the sport's physical demands, which can be used to inform training programme design. [ABSTRACT FROM AUTHOR]