Citation

De la Fuente C, Silvestre R, Yañez R, Roby M, Soldán M, Ferrada W, Carpes FP. Sci. Med. Footb. 2022; ePub(ePub): ePub.

Copyright

(Copyright © 2022, Informa - Taylor and Francis Group)

DOI

10.1080/24733938.2022.2075558

PMID

35522903

Abstract

Injury risk is regularly assessed during the preseason in susceptible populations like female soccer players. However, multiple outcomes (high-dimensional dataset) derived from multiple testing may make pattern recognition difficult. Thus, dimension reduction and clustering may be useful for improving injury surveillance when results of multiple assessments tools are available. Thus, we determined the influence of dimension reduction for pattern recognition followed by clustering on multiple biomechanical injury markers in elite female soccer players during preseason. We introduce the use of dimension reduction through linear principal component analysis (PCA), non-linear kernel principal component analysis (k-PCA), t-distributed stochastic neighbor embedding (t-sne), and uniform manifold approximation and projection (umap) for injury markers via grid search. Muscle strength, muscle function, jump technique and power, balance, muscle stiffness, exercise tolerance, and running performance were assessed in an elite female soccer team (n=21) prior to the competitive season. As a result, umap facilitated the injury pattern recognition compared to PCA, k-PCA, and t-sne. One of three patterns was related to a team subgroup with acceptable muscle conditions. In contrast, the other two patterns showed higher injury risk profiles. For our dataset, umap improved injury surveillance through multiple testing characteristics. Dimension reduction and clustering techniques present as useful strategies to analyze subgroups of female soccer players who have different risk profiles for injury.

Language: en

Keywords

Sports; Machine learning; Football; Biomechanics; Non-linear reduction