Validation and Reliability of a Classification Method to Measure the Time Spent Performing Different Activities

• Published in: PloS one 2015-06, Vol.10 (6), p.e0128299-e0128299
• Main Authors: Riou, Marie-Ève, Rioux, François, Lamothe, Gilles, Doucet, Éric
• Format: Article
• Language: English
• Subjects: Accelerometers; Accelerometry - methods; Adult; Artificial intelligence; Classification; Experiments; Female; Health sciences; Humans; Male; Metabolism; Model testing; Models, Theoretical; Monitoring systems; Motor Activity; Obesity; Reliability; Reproducibility; Reproducibility of Results; Sensors; Thigh; Time Factors; Time measurement; Validation studies; Walking; Young Adult
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0128299

The aim of this study was to validate the performance and reliability of results obtained from a classification model that measures time spent performing activities in confined (CE) and unrestricted (UE) environments. In CE, participants wore a pair of biaxial and/or triaxial accelerometers while performing pre-determined training activities classified as variants of lying down, dynamic standing, sitting, walking and running on two separate days. A classification model trained with activities performed in a specific order during the first day was developed to validate the activities performed in a random order on the second day (CE) and over 24 hours on a separate day (UE). The performance of the classification model was validated against triaxial accelerometers using six (x, y and step counts for arm and thigh) or eight (same as six features plus z axis) features. The reliability of the classification model was tested in both environments using six features. Results revealed an overall accuracy of 94% in CE and 90% in UE. The sensitivity in CE and UE was 94% and 95% for lying down, 88% and 80% for dynamic standing, 97% and 89% for sitting, 96% and 78% for walking and 90% and 64% for running, respectively. No significant differences were noted between performances obtained with six or eight features. Results were highly reproducible in both environments. The results obtained from the classification model were accurate and reproducible, and highlight the potential use of this approach in research to quantify the time spent performing different activities.