Invited Speaker ANZOS Annual Scientific Meeting 2021

Device-based measures of movement behaviour. Is it time to say goodbye to cut-point methods? (#47)

Stewart Trost 1
  1. Queensland University of Technology, South Brisbane, QLD, Australia

Accelerometer-based monitoring of movement behaviours has become popular among researchers and consumers. However, the research potential of accelerometers has been severely under-utilised, with analysis restricted to the application of simple cut-points or linear regression models based on proprietary activity counts. Machine learning approaches to accelerometer data reduction have emerged as a more accurate and versatile alternative to cut-point methods. However, the uptake of machine learning methods by clinical and public health researchers has been limited, in part, due to the difficulties of implementation, the concern that models trained on data from laboratory-based activity trials do not generalize well to free-living environments, and the lack of studies demonstrating the relative advantage of machine learning approaches over traditional cut-point methods. This presentation will focus on the application of machine learning accelerometer data processing methodologies for human activity recognition. It will demonstrate the advantages of machine learning methods relative to cut-point methods and introduce deployment tools that enable health researchers without specialist training in data science to implement machine learning physical activity classification and energy expenditure estimation algorithms.