Author(s): DEGHAIES, K., LUSSIANA, T., TOUZARD, P., FOUREL, L., OZAN, S., GINDRE, C., BIDEAU, B., MARTIN, C., Institution: M2S LABORATORY RENNES 2 UNIVERSITY, Country: FRANCE, Abstract-ID: 2435

INTRODUCTION:
The tennis serve can be divided into two dynamic distinct phases – loading and acceleration. Here we focus on the loading phase which may reflect style and individual tendency rather than substance (Kovacs & Ellenbecker, 2011) and where lower limb’s inter-individual variability seems to be important (Fleisig et al., 2003). Contemporary approaches for profiling sport-specific movements are currently emerging and being utilized in sports biomechanics. This study aims to employ one of these approaches (clustering analysis) to objectively identify potential serve motion strategies among professional tennis players and compare serve kinematics and performance indicators across these strategies.
METHODS:
Thirty-one male ATP ranked tennis players, using a foot-up serve technique, performed 5 successful flat serves in a target area. Three-dimensional marker trajectories were recorded at 300 Hz using a 23-camera Qualisys motion analysis system. A radar was positioned behind the baseline to measure ball speed. We performed a two-step cluster analysis. First, hierarchical clustering with Ward’s method and squared Euclidean distance and then a k-means clustering in order to classify players’ strategy according to several kinematic variables of the loading phase: downward center of mass (CoM) range of motion, maximum downward CoM velocity, and loading phase duration. The clustering models quality was assessed using the average silhouette coefficient, measuring both cluster cohesion and separation. Performance indicators (ball speed and impact height) and serve kinematics parameters were compared between clusters.
RESULTS:
Two clusters (CL1: n = 17, CL2: n = 14) were identified with an average silhouette coefficient of 0.37 indicating a moderate to fair clustering model. No significant difference was found between clusters in height, weight, age, ATP ranking, impact height, or ball speed. Players of CL1 showed greater downward CoM range of motion (-7.87 ± 1.45 vs. -3.77 ± 1.31% of body height, p < 0.001), as well as greater maximum downward CoM velocity (-0.45 ± 0.13 vs. -0.25 ± 0.08 m.s-1, p < 0.001). However, players of CL2 showed greater horizontal maximal velocity during the loading phase (1.04 ± 0.24 vs. 0.76 ± 0.16 vs. m.s-1, p < 0.001). Moreover, the start of the loading phase relative to the instant of ball release (BR) (CL1: 78 ± 78 ms after BR, CL2: 39 ± 127 ms prior to BR, p = 0.004), as well as the duration between BR and ball-impact (CL1: 1032 ± 95 ms, CL2: 888 ± 170 ms, p = 0.006) were significantly different between the two clusters.
CONCLUSION:
The observed differences between clusters highlight two distinct effective serve strategies in professional players. One is characterized by a higher, later, and faster downward CoM range of motion, while the other is characterized by an earlier and lower downward CoM range of motion, and a higher horizontal CoM velocity during the loading phase. Coaches should consider both these strategies to better individualize serve instructions.