Development and Validation of a Machine Learning Wrist-worn Step Detection Algorithm with Deployment in the UK Biobank

Abstract

We have developed a new open-source step counting method, informed by self-supervised machine learning methods that substantially outperforms current wrist-worn step counting algorithms in the free-living environment. The open data and code released with this manuscript will provide the global research community access to a more transparent and well-validated method to measure steps in large-scale wrist-worn accelerometer datasets. When applying the algorithm and resulting step metric in epidemiological analysis, we demonstrated that a higher daily step count is associated with a lower risk of