TY - JOUR
T1 - A novel algorithm to detect non-wear time from raw accelerometer data using deep convolutional neural networks
AU - Syed, Shaheen
AU - Morseth, Bente
AU - Hopstock, Laila A.
AU - Horsch, Alexander
N1 - Publisher Copyright:
© 2021, The Author(s).
PY - 2021/12
Y1 - 2021/12
N2 - To date, non-wear detection algorithms commonly employ a 30, 60, or even 90 mins interval or window in which acceleration values need to be below a threshold value. A major drawback of such intervals is that they need to be long enough to prevent false positives (type I errors), while short enough to prevent false negatives (type II errors), which limits detecting both short and longer episodes of non-wear time. In this paper, we propose a novel non-wear detection algorithm that eliminates the need for an interval. Rather than inspecting acceleration within intervals, we explore acceleration right before and right after an episode of non-wear time. We trained a deep convolutional neural network that was able to infer non-wear time by detecting when the accelerometer was removed and when it was placed back on again. We evaluate our algorithm against several baseline and existing non-wear algorithms, and our algorithm achieves a perfect precision, a recall of 0.9962, and an F1 score of 0.9981, outperforming all evaluated algorithms. Although our algorithm was developed using patterns learned from a hip-worn accelerometer, we propose algorithmic steps that can easily be applied to a wrist-worn accelerometer and a retrained classification model.
AB - To date, non-wear detection algorithms commonly employ a 30, 60, or even 90 mins interval or window in which acceleration values need to be below a threshold value. A major drawback of such intervals is that they need to be long enough to prevent false positives (type I errors), while short enough to prevent false negatives (type II errors), which limits detecting both short and longer episodes of non-wear time. In this paper, we propose a novel non-wear detection algorithm that eliminates the need for an interval. Rather than inspecting acceleration within intervals, we explore acceleration right before and right after an episode of non-wear time. We trained a deep convolutional neural network that was able to infer non-wear time by detecting when the accelerometer was removed and when it was placed back on again. We evaluate our algorithm against several baseline and existing non-wear algorithms, and our algorithm achieves a perfect precision, a recall of 0.9962, and an F1 score of 0.9981, outperforming all evaluated algorithms. Although our algorithm was developed using patterns learned from a hip-worn accelerometer, we propose algorithmic steps that can easily be applied to a wrist-worn accelerometer and a retrained classification model.
UR - http://www.scopus.com/inward/record.url?scp=85104856666&partnerID=8YFLogxK
U2 - 10.1038/s41598-021-87757-z
DO - 10.1038/s41598-021-87757-z
M3 - Article
C2 - 33893345
AN - SCOPUS:85104856666
SN - 2045-2322
VL - 11
JO - Scientific Reports
JF - Scientific Reports
IS - 1
M1 - 8832
ER -