Unsupervised Early Detection of Physical Activity Behaviour Changes from Wearable Accelerometer Data

Abstract

Wearable accelerometers record physical activity with high resolution, potentially capturing the rich details of behaviour changes and habits. Detecting these changes as they emerge is valuable information for any strategy that promotes physical activity and teaches healthy behaviours or habits. Indeed, this offers the opportunity to provide timely feedback and to tailor programmes to each participant’s needs, thus helping to promote the adherence to and the effectiveness of the intervention. This article presents and illustrates U-BEHAVED, an unsupervised algorithm that periodically scans step data streamed from activity trackers to detect physical activity behaviour changes to assess whether they may become habitual patterns. Using rolling time windows, current behaviours are compared with recent previous ones, identifying any significant change. If sustained over time, these new behaviours are classified as potentially new habits. We validated this detection algorithm using a physical activity tracker step dataset (N = 12,798) from 79 users. The algorithm detected 80% of behaviour changes of at least 400 steps within the same hour in users with low variability in physical activity, and of 1600 steps in those with high variability. Based on a threshold cadence of approximately 100 steps per minute for standard walking pace, this number of steps would suggest approximately 4 and 16 min of physical activity at moderate-to-vigorous intensity, respectively. The detection rate for new habits was 80% with a minimum threshold of 500 or 1600 steps within the same hour in users with low or high variability, respectively.

Patterns of physical activity and sedentary behavior in child and adolescent cancer survivors assessed using wrist accelerometry: A cluster analysis approach

“…A bout is defined as continuous physical activity at a certain intensity level. 27 The length of a bout is defined as the number of seconds spent during that period. For example, a participant spending 10 seconds in moderate intensity physical activity followed by 4 seconds of vigorous physical activity, then 10 seconds of light activity would generate one bout of moderate-to-vigorous physical activity (MVPA) lasting 14 seconds.…”

“…25 We used the thresholds of 3, 10, and 30 s for MVPA and 60, 120, and 300 s for sedentary bouts based on a previous study that showed a high dispersion between children’s cumulative sum of physical activity levels. 27 Separating the physical activity bouts by these thresholds allowed us to create meaningful features and generate cohesive clusters. After data normalisation and using the Principal Components Analysis to maximise the variance of our data, we retained three principal components and applied the k-means unsupervised algorithm with k = 5 for the daily movement clusters and k = 4 for hourly movement clusters.…”

“…24 We extracted the binary files as data frames into triaxial acceleration records with timestamps for every participant to calculate gravity-subtracted Signal Vector Magnitudes (SVMgs), with gravity estimated to 1 for each one second epoch. 16,25 We applied one second epochs as children typically achieve their physical activity in short bursts. 25 Next, we grouped each second of data of SVMgs into a physical activity intensity level using cut-offs that are validated for the assessment of physical activity in children.…”

“…13 The most frequent use of accelerometers among survivors has been to quantify physical activity in each intensity level (i.e., light, moderate, vigorous) however, this single evaluation of total physical activity does not consider the changing patterns of physical activity and sedentary behaviors over the course of a 24-hr day. 16 Children and adolescents have a transient nature of physical activity and typically engage in intermittent and sporadic activity, rather than continuous. 17 Understanding the temporal patterns and variation of physical activity and sedentary behaviors among survivors may enhance our knowledge on strategies to improve health behaviors in this cohort.…”

“…Data mining analyses may help to close the gap of fragmented, unreliable, or inaccurate self-reported activity and will ultimately improve our understanding of how to engage survivors to move more throughout the day. 19 Data mining techniques have been used in several populations including children and adolescents without chronic conditions, 16 and adults with chronic conditions such as heart disease 20 and diabetes. 21 However, this technique has been rarely applied in children who have survived cancer.…”