CLUSTERING OF RUNNERS BASED ON RUNNING STYLE AND COORDINATION
Authors
Abstract
This thesis evaluates six non-linear dimensionality reduction techniques ¿ Isomap, LLE, MDS, Spectral Embedding, t-SNE and UMAP ¿ applied to a dataset of 1267 runners combining kinematic time series and demographic metadata. A scoring system based on five metrics (Silhouette, Davies-Bouldin, Trustworthiness, R² kNN and Spearman) was developed to identify the best algorithm and variable combination. Results show that runner speed is the variable best captured by the embeddings, with UMAP achieving the highest score (0.703). A permutation analysis revealed that the kinematic data of the ankle, foot and hip joints carry the most weight in the embedding structure, providing a biomechanical explanation consistent with the literature.
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Rey Juan Carlos
URL external
External URL
DOI
Date
Description
Trabajo Fin de Grado leído en la Universidad Rey Juan Carlos en el curso académico 2026/2027. Directores/as: Francisco Manuel Melgarejo Meseguer, Sara García De Villa

