Application of Convolutional Neural Network Algorithms for Advancing Sedentary and Activity Bout Classification

Abstract

Background: Machine learning has been used for classification of physical behavior bouts from hip-worn accelerometers; however, this research has been limited due to the challenges of directly observing and coding human behavior “in the wild.” Deep learning algorithms, such as convolutional neural networks (CNNs), may offer better representation of data than other machine learning algorithms without the need for engineered features and may be better suited to dealing with free-living data. The purpose of this study was to develop a modeling pipeline for evaluation of a CNN model on a free-living data set and compare CNN inputs and results with the commonly used machine learning random forest and logistic regression algorithms. Method: Twenty-eight free-living women wore an ActiGraph GT3X+ accelerometer on their right hip for 7 days. A concurrently worn thigh-mounted activPAL device captured ground truth activity labels. The authors evaluated logistic regression, random forest, and CNN models for classifying sitting, standing, and stepping bouts. The authors also assessed the benefit of performing feature engineering for this task. Results: The CNN classifier performed best (average balanced accuracy for bout classification of sitting, standing, and stepping was 84%) compared with the other methods (56% for logistic regression and 76% for random forest), even without performing any feature engineering. Conclusion: Using the recent advancements in deep neural networks, the authors showed that a CNN model can outperform other methods even without feature engineering. This has important implications for both the model’s ability to deal with the complexity of free-living data and its potential transferability to new populations.

Physical activity intensity profiles associated with cardiometabolic risk in middle-aged to older men and women

“…Future research utilising this method, alongside thigh-worn devices or algorithms (e.g. deep learning) better equipped to distinguish posture ( Aguilar-Farías et al, 2014 ; Jain et al, 2021 ; Nakandala et al, 2021 ), or the use of raw acceleration measurement coupled with low-frequency extension filtering for greater sensitivity to low-intensity activities of older-aged sub-groups ( Cain et al, 2013 ), could help interrogate these aspects further.…”

Agreement of Sedentary Behavior Metrics Derived From Hip- and Thigh-Worn Accelerometers Among Older Adults: With Implications for Studying Physical and Cognitive Health

“…Future development of sedentary behavior guidelines will rely on accelerometer data from longitudinal studies. As many studies have already collected hip-worn AG data and many new studies will likely continue to employ the AG, research is needed to develop accelerometer data processing techniques (Kerr et al, 2018;Nakandala et al, 2021) or measurement error correction models (Sampson, Matthews, Freedman, Carroll, & Kipnis, 2016) to increase the accuracy of sedentary behavior pattern metrics derived from hip-worn devices. Until then, for studies assessing associations of sedentary behavior volumes and patterns with health using AG 100cpm , we recommend presenting results for standardized and unstandardized point estimates in AG 100cpmbased studies and interpreting associations using units of the standardized metric rather than absolute units of sedentary behavior measures (e.g., minute, minute/day, n/day).…”

Cohort profile: the Women’s Health Accelerometry Collaboration

“…Raw accelerometry data will allow the research team to develop further measures of physical activity and sedentary behaviour, such as using the activity index 32 and latent class analysis on accelerometry. 33 Using the raw data, we can also apply two machine-learnt algorithms developed specifically for older women; one designed to distinguish sitting, riding in a vehicle, standing still, standing moving and walking, 34 while the other was designed to accurately quantify sitting bouts, 35 which, without the algorithm, are measured with substantial error. 36 While studies investigating the associations between less common cancer subtypes and physical activity or sedentary behaviour among older women have been limited due to smaller sample sizes and few cancer events, the combined cohorts provide improvement in statistical power, allowing researchers to be better equipped to investigate these associations.…”