Categorizing race profiles in professional men’s and women’s road cycling races using clustering

Abstract

Abstract Background Professional road cycling races differ in physiological demands because race profiles vary. Yet no standardized method exists to categorize race profiles across professional road cycling races. Organizer labels are inconsistent or missing. Systematic categorization would enable large-scale analysis of race demands. This study develops a categorization of race profiles in men’s and women’s professional road cycling. Methods The dataset comprised international men’s and women’s races from 2024 supplemented with women’s stage races from 2023. Races with an organizer-assigned profile labels were included (men: 262; women: 128). Features from physiological-focused studies (distance, vertical meters, number of climbs, uphill distance, vertical meters highest climb, vertical meters climbs in the last 10 km, and at the finish) were extracted from ProcyclingStats. KMeans, hierarchical clustering with Genie linkage (Genie), and Ordering Points To Identify Cluster Structure (OPTICS) were compared, selecting the highest-Silhouette configuration per algorithm. The larger men’s dataset was used to evaluate clustering performance against a subset of organizer-provided labels serving as reference label. Each cluster was matched to its most frequently occurring label, and accuracy calculated as the percentage of matching races. The best-performing algorithm was then applied to women’s data. Results In men's races, KMeans and Genie each produced five clusters (Silhouette: 0.43 and 0.49), while OPTICS produced three (0.83). Against organizer-labels, KMeans achieved the highest accuracy (62% vs 46% for Genie and 3% for OPTICS), identifying: flat, hilly, semi-mountain, mountain and time trial. Applying KMeans to women's races yielded eight clusters, one flat, two hilly, 2 semi-mountain, 2 mountain and time trial (Silhouette: 0.48, accuracy: 63%). Conclusions This study proposes a framework for categorization race profiles in men’s and women’s cycling professional road cycling. The categorization may facilitate large-scale systematic analysis to gain more precise insights into the specific demands of each profile, which could help align a cyclist’s strengths with specific race demands, thereby informing rider and race selection and may guide training goals and race periodisation. All of which may contribute to performance optimization. Further research is needed to validate the framework against an independent and well-established reference standard.