Identifying cycling behaviour in healthy adults using thigh-worn accelerometry and activity classification algorithms

Claas Lendt*, Peter Johansson, Theresa Braun, Bianca Biallas

*Korrespondierende*r Autor*in für diese Arbeit

Publikation: Beitrag in FachzeitschriftKonferenz-Abstract in FachzeitschriftForschungBegutachtung

Abstract

Introduction: Cycling is associated with reduced mortality and morbidity. Moreover, an increasing number of health promotion efforts aim to increase cycling as part of active commuting. Accurate methods to identify cycling are crucial to advance our understanding of associated health benefits and overall cycling behaviour. Thigh-worn accelerometers can be used to objectively determine the duration and frequency of basic physical activity types performed over several days. Previous research suggests that the classification accuracy of free-living cycling remains challenging, but no study has yet evaluated differences between available algorithms. In this study, we compare two classification algorithms for thigh-worn accelerometer data regarding their accuracy in correctly classifying cycling.
Methods: 35 healthy adults (51% female, age = 30.1± 9.0 years, BMI = 23.6± 3.1 kg/m²) were equipped with a SENS motion triaxial accelerometer (12.5 Hz with ± 4g) attached to the lateral thigh. Participants performed a standardised 3x3-minute laboratory protocol on a cycling ergometer with varying intensities, followed by 60 minutes of unrestricted free-living activities with a video camera mounted to the chest. Time-synchronised videos were used to annotate the start and end of cycling. Raw accelerometer data was processed using the SENS motion web-application and ActiPASS Version 1.58.
Results: ActiPASS and SENS motion algorithms both achieved a balanced accuracy of 0.96 for cycling under laboratory conditions. A total of 297 minutes of cycling was annotated for the free-living condition. ActiPASS achieved a balanced accuracy of 0.90 (sensitivity = 0.82; specificity = 0.99) for free-living cycling while SENS motion achieved a balanced accuracy of 0.80 (sensitivity = 0.59; specificity = 1.00). Post-processing of the free-living SENS motion classification using a moving majority voting algorithm improved the balanced accuracy to 0.86 and sensitivity to 0.72.
Conclusions: Researchers may use thigh-worn accelerometers and existing algorithms to objectively identify free-living cycling behaviour with high accuracy. The ActiPASS classification algorithm performed more accurate than the SENS motion algorithm. Post-processing techniques such as filtering can potentially improve the classification sensitivity and contribute to more accurate classifications.
OriginalspracheEnglisch
ZeitschriftThe Journal of Sport and Exercise Science
Jahrgang7
Ausgabenummer4
Seiten (von - bis)86-87
Seitenumfang2
ISSN2703-240X
DOIs
PublikationsstatusVeröffentlicht - 20.11.2023
VeranstaltungAsia-Pacific Society for Physical Activity (ASPA) / Sport & Exercise Science New Zealand (SESNZ) Annual Conference - Te Herenga Waka – Victoria University of Wellington, Wellington, Neuseeland
Dauer: 27.11.202329.11.2023
https://sesnz.org.nz/conference/

Fingerprint

Untersuchen Sie die Forschungsthemen von „Identifying cycling behaviour in healthy adults using thigh-worn accelerometry and activity classification algorithms“. Zusammen bilden sie einen einzigartigen Fingerprint.

Zitation