Self-supervised learning for human activity recognition using 700,000 person-days of wearable data

Accurate physical activity monitoring is essential to understand the impact of physical activity on one's physical health and overall well-being. However, advances in human activity recognition algorithms have been constrained by the limited availability of large labelled datasets. This study aims to leverage recent advances in self-supervised learning to exploit the large-scale UK Biobank accelerometer dataset-a 700,000 person-days unlabelled dataset-in order to build models with vastly improved generalisability and accuracy. Our resulting models consistently outperform strong baselines across eight benchmark datasets, with an F1 relative improvement of 2.5-130.9% (median 24.4%). More importantly, in contrast…

• G
  GlaxoSmithKline
• W
  Wellcome

  Award: [223100/Z/21/Z]
• WT
  Wellcome Trust

  Award: 223100/Z/21/Z
• UR
  UK Research and Innovation
• NI
  National Institute for Health and Care Research
• BH
  British Heart Foundation

  Award: RE/18/3/34214
• DO
  Department of Health and Social Care
• RA
  Royal Academy of Engineering
• UO
  University of Exeter
• UO
  University of Oxford
• NN
  Novo Nordisk
• S
  Samsung
• EA
  Engineering and Physical Sciences Research Council

  Awards: EP/S001530/1, EP/R018677/1, EP/R018677/1, EP/S001530