Methods / Statistics
The aim of this study was to assess whether clustering runners based on their technique resulted in consistent group allocations across multiple speeds. Eighty-four runners (34 females) completed four 4-minute running stages at 10, 11, 12 and 13 km/h. For each stage, running technique was characterised using a set of continuous variables in the sagittal plane and discrete stride-based variables. An autoencoder neural network was used for dimensionality reduction and agglomerative hierarchical clustering was applied to identify groups of runners with a similar technique. Two clusters for each speed were selected and the clustering partitions at different incremental speeds were compared. Our results showed that partitions were inconsistent across speeds, and therefore clustering results at one single speed do not generalise to the range of speeds an athlete typically runs at. Single speed clustering may be limited to drive the design of cluster-specific running training interventions and different clustering approaches are needed to better capture runners’ technique at their typical speeds.
New Investigator Award
Rivadulla, Adrian R.; Chen, Xi; Cazzola, Dario; Trewartha, Grant; and Preatoni, Ezio
"CLUSTERING LONG-DISTANCE RUNNERS BASED ON THEIR TECHNIQUE AT ONE SINGLE SPEED DOES NOT GENERALISE TO MULTIPLE SPEEDS,"
ISBS Proceedings Archive: Vol. 41:
1, Article 91.
Available at: https://commons.nmu.edu/isbs/vol41/iss1/91