Logistic regression with elastic net regularisation can successfully identify groups of runners based on their technique using consumer-oriented wearable sensor data

We previously identified that runners can be clustered in two main groups based on their technique: the “neutral” and “tilted pelvis” clusters. Cluster-specific interventions may enhance training success, but identifying what cluster an athlete belongs to currently requires a lab-based optical motion capture system. Here, we develop and validate a regularised logistic regression model that uses data from consumer-oriented wearable technologies to allocate runners to one of the two running technique classes. Using 2-3 convenient sensor locations was enough to achieve testing scores ≥ 0.82, enabling reasonably confident allocation of new runners to the neutral and tilted pelvis techniques. This method facilitates large scale cluster-specific training development and provides real-world solutions for the wider running community.