Unsupervised phenotypic clustering, inferential characterization, and discriminant classification of elite flatwater canoeists: neuromuscular and anthropometric determinants
- 1Department of Physical Education and Sports, Faculty of Sports Sciences and Physical Education, University of A Coruña , ,
- 2Sport Performance, Physical Condition and Wellness lab, Faculty of Education and Sports Sciences, University of Vigo , ,
- 3Department of Design in Engineering, School of Industrial Engineering, University of Vigo , ,
Abstract
Introduction: Phenotypic profiling of elite athletes remains constrained by supervised frameworks that conflate physiological characteristics with performance outcomes. Unsupervised machine learning offers an outcome-independent alternative for identifying intrinsic athlete profiles.
Method: This study applied clustering algorithms to comprehensive laboratory data from elite flatwater canoeists. Twenty-one international-level C1 canoeists (12 male, 9 female) underwent a standardized testing battery including body composition, muscle contractile properties, mobility, maximal and explosive strength, and sport-specific neuromuscular assessments via Functional Electromechanical Dynamometry (FEMD). All performance variables were excluded to ensure purely phenotypic segmentation. Preprocessing included asymmetric winsorization, Yeo-Johnson transformation, and robust scaling. K-Means and hierarchical clustering (Ward linkage) were applied with optimal k determined via silhouette coefficient, Davies-Bouldin index, and elbow method. Discriminant variable importance was assessed via Random Forest Classifier; interpretable classification rules were extracted using Decision Tree analysis (max_depth=3) with stratified train-test validation (66.7%–33.3%). Statistical characterization of cluster differences employed ANOVA or Kruskal-Wallis tests with post-hoc comparisons and Benjamini-Hochberg FDR correction; algorithmic concordance was quantified via Adjusted Rand Index (ARI).
Results: Three phenotypic clusters emerged: High-Neuromuscular Male Phenotype (n=7, 33.3%), Female-Dominant Phenotype (n=10, 47.6%), and Moderate-Neuromuscular Male Phenotype (n=4, 19.0%) (ARI = 0.591). Sex stratification was marked (χ²=18.90, p<0.001): the Female-Dominant Phenotype comprised 90% female athletes; both male phenotypes were exclusively male, encoding sexual dimorphism alongside intra-sex neuromuscular differentiation. FEMD neuromuscular variables dominated discrimination (75.9% of top-15 importance; 14 of 15 variables with padj < 0.01 following Benjamini-Hochberg FDR correction). Decision Tree classification achieved 100% training accuracy and 71.4% test accuracy (5/7; 95% CI: 29.0%–96.3%; p=0.045 vs. chance), though the small test partition constrains generalizability. Within male phenotypes, Clusters 0 and 2 demonstrated equivalent neuromuscular profiles (all p > 0.05), with phenotypic separation captured by latissimus dorsi sustain time from the non-paddle side; differences relative to the Female-Dominant Phenotype primarily reflect sexual dimorphism.
Discussion: Unsupervised clustering suggests a tripartite phenotypic structure in elite canoeists consistent with sexual dimorphism and intra-sex neuromuscular differentiation, with FEMD-derived eccentric force as primary discriminant. Classification rules may inform future sex-stratified talent identification and training individualization; external validation with larger cohorts is required.
Related
- An integrated wireless sEMG framework with Bayesian-optimized machine learning for multi-task locomotor classification and neuromuscular characterization
- An innovative framework for student adaptability prediction using discriminant-based classification, feature selection, and sensitivity analysis
- Advancing Heart Failure Classification Beyond Left Ventricular Ejection Fraction: Unsupervised Machine Learning Insights From the Japanese Nationwide Acute Decompensated Heart Failure Cohort: JROADHF
- Unsupervised Classification of Wide-Field Swept-Source Optical Coherence Tomography Reveals Patterns Associated With Glaucoma Severity and Progression.
- Unsupervised learning on advanced diffusion MRI metrics for clinical characterization of migraine
BibTeX
@article{alvarezyates2026unsupervised,
title = {{Unsupervised phenotypic clustering, inferential characterization, and discriminant classification of elite flatwater canoeists: neuromuscular and anthropometric determinants}},
author = {Tania Álvarez-Yates and Óscar García-García and Alba Cuba-Dorado and José-Benito Bouza-Rodríguez and Alberto Comesaña-Campos and Jorge Cerqueiro-Pequeño},
journal = {Frontiers in physiology},
year = {2026},
month = oct,
doi = {10.3389/fphys.2026.1949031},
url = {https://europepmc.org/article/PMC/PMC13725060},
note = {CC BY},
}
RIS
TY - JOUR
TI - Unsupervised phenotypic clustering, inferential characterization, and discriminant classification of elite flatwater canoeists: neuromuscular and anthropometric determinants
AU - Tania Álvarez-Yates
AU - Óscar García-García
AU - Alba Cuba-Dorado
AU - José-Benito Bouza-Rodríguez
AU - Alberto Comesaña-Campos
AU - Jorge Cerqueiro-Pequeño
PY - 2026
DA - 2026/10/19
JO - Frontiers in physiology
DO - 10.3389/fphys.2026.1949031
UR - https://europepmc.org/article/PMC/PMC13725060
AB - Introduction: Phenotypic profiling of elite athletes remains constrained by supervised frameworks that conflate physiological characteristics with performance outcomes. Unsupervised machine learning offers an outcome-independent alternative for identifying intrinsic athlete profiles. Method: This study applied clustering algorithms to comprehensive laboratory data from elite flatwater canoeists. Twenty-one international-level C1 canoeists (12 male, 9 female) underwent a standardized testing battery including body composition, muscle contractile properties, mobility, maximal and explosive strength, and sport-specific neuromuscular assessments via Functional Electromechanical Dynamometry (FEMD). All performance variables were excluded to ensure purely phenotypic segmentation. Preprocessing included asymmetric winsorization, Yeo-Johnson transformation, and robust scaling. K-Means and hierarchical clustering (Ward linkage) were applied with optimal k determined via silhouette coefficient, Davies-Bouldin index, and elbow method. Discriminant variable importance was assessed via Random Forest Classifier; interpretable classification rules were extracted using Decision Tree analysis (max_depth=3) with stratified train-test validation (66.7%–33.3%). Statistical characterization of cluster differences employed ANOVA or Kruskal-Wallis tests with post-hoc comparisons and Benjamini-Hochberg FDR correction; algorithmic concordance was quantified via Adjusted Rand Index (ARI). Results: Three phenotypic clusters emerged: High-Neuromuscular Male Phenotype (n=7, 33.3%), Female-Dominant Phenotype (n=10, 47.6%), and Moderate-Neuromuscular Male Phenotype (n=4, 19.0%) (ARI = 0.591). Sex stratification was marked (χ²=18.90, p<0.001): the Female-Dominant Phenotype comprised 90% female athletes; both male phenotypes were exclusively male, encoding sexual dimorphism alongside intra-sex neuromuscular differentiation. FEMD neuromuscular variables dominated discrimination (75.9% of top-15 importance; 14 of 15 variables with padj < 0.01 following Benjamini-Hochberg FDR correction). Decision Tree classification achieved 100% training accuracy and 71.4% test accuracy (5/7; 95% CI: 29.0%–96.3%; p=0.045 vs. chance), though the small test partition constrains generalizability. Within male phenotypes, Clusters 0 and 2 demonstrated equivalent neuromuscular profiles (all p > 0.05), with phenotypic separation captured by latissimus dorsi sustain time from the non-paddle side; differences relative to the Female-Dominant Phenotype primarily reflect sexual dimorphism. Discussion: Unsupervised clustering suggests a tripartite phenotypic structure in elite canoeists consistent with sexual dimorphism and intra-sex neuromuscular differentiation, with FEMD-derived eccentric force as primary discriminant. Classification rules may inform future sex-stratified talent identification and training individualization; external validation with larger cohorts is required.
SN - 1664-042X
ER -