Deep CHORES: Estimating Hallmark Measures of Physical Activity Using Deep Learning.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Human Movement Science

Background:

• Smartwatch technology has increased the use of wrist accelerometers for physical activity (PA) assessment.
• A need exists for rigorous evaluation of these devices for recognizing PA type and estimating energy expenditure (EE) across diverse age groups.

Purpose of the Study:

• To evaluate the performance of deep learning models in recognizing PA type and estimating EE using wrist-worn tri-axial accelerometer data.
• To assess the accuracy of these models across a wide age range (20-89 years).

Main Methods:

• Participants (N=unknown, 66% women) performed 33 standardized daily activities.
• Tri-axial accelerometer data from the wrist and metabolic unit data were collected.
• Deep learning networks were developed to analyze time-series data for PA recognition and EE estimation.

Main Results:

• Deep learning models achieved high accuracy in PA recognition: F1 scores of 0.82 (sedentary), 0.81 (locomotor), and 0.95 (lifestyle activities).
• Energy expenditure (EE) estimation showed a root mean square error of 1.1 (+/-0.13).

Conclusions:

• Deep learning models demonstrate high efficacy in classifying physical activity types and estimating energy expenditure from wrist accelerometer data.
• These findings support the potential of wrist-worn accelerometers and advanced algorithms for objective PA and EE assessment in population health studies.