Clustering tennis players’ anthropometric and individual features helps to reveal performance fingerprints

Abstract

The study was aimed to explore distinct players’ groups according to their anthropometric and individual features, and to identify the key performance indicators that discriminate player groups. Match statistics, anthropometric and personal features of 1188 male players competing during 2015–2017 main draw Grand Slam singles events were collected. Height, weight, experience, handedness and backhand style were used to automatically classify players into different clusters through unsupervised learning model. Afterwards, 29 match variables were analysed through MANOVA and discriminant analysis in order to evaluate the different match performance among player groups and to identify the key performance indicators that best differentiate player clusters in each Grand Slam. The analysis revealed the existence of four clusters, they were classified as Big-sized Right Two-handed Players (n = 387), Medium-sized Right One-handed Players (n = 265), Small-sized Right Two-handed Players (n = 414), and Left Two-handed Players (n = 122). Serve, winner, net and physical performance-related indicators (Structure Coefficient ≥ |0.30|) were showed to be the maximum contributors to the group separation. Left-handed players were the most homogenous group in performance. Taller players outperformed their peers in all Slams except for Roland Garros, where left-handed players demonstrated certain advantage playing on slow-pace surface. In Wimbledon and US Open, Medium-sized Right One-handed Players showed better net and physical performance. The advantage of left-handed player is over-represented at elite level. Current findings promote a better understanding of match-play from distinct player groups and offer information on evaluating contextual variability for achieving better performances.

Keywords:

• Professional tennis
• player groups
• cluster analysis
• key performance

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data for this article can be accessed here (https://doi.org/10.1080/17461391.2019.1577494).

ORCID

Yixiong Cui http://orcid.org/0000-0002-1755-9631

Miguel-Ángel Gómez http://orcid.org/0000-0002-9585-3158

Bruno Gonçalves http://orcid.org/0000-0001-7874-4104

Jaime Sampaio http://orcid.org/0000-0003-2335-9991

Additional information

Funding

The first author was supported by the China Postdoctoral Science Foundation. CIDESD is a research unit supported by the Portuguese Foundation for Science and Technology (UID/DTP/04045/2013) and by the European Regional Development Fund, through COMPETE 2020 (POCI-01-0145-FEDER-006969). This research is part of NanoSTIMA: Macro-to-Nano Human Sensing: Towards Integrated Multimodal Health Monitoring and Analytics, funded by the European Regional Development Fund, through NORTE 2020 (NORTE-01-0145-FEDER-000016).