Composite activity type and stride-specific energy expenditure estimation model for thigh-worn accelerometry.

Area of Science:

• Biomedical Engineering
• Sports Science
• Wearable Technology

Background:

• Accurate measurement of energy expenditure during physical activity outside laboratory settings is challenging.
• Thigh-worn accelerometers offer potential for large-scale physical activity monitoring.
• Machine learning (ML) can enhance activity classification and energy expenditure prediction accuracy.

Purpose of the Study:

• Develop a novel composite energy expenditure estimation model.
• Combine an activity classification model with a stride-specific energy expenditure model.
• Improve accuracy of energy expenditure measurements in non-laboratory settings.

Main Methods:

• Trained a deep learning activity classification model using pooled adult accelerometer data.
• Developed and validated a composite energy expenditure model with 69 healthy adults.
• Used indirect calorimetry as the reference measure for validation.

Main Results:

• Activity classification model achieved 99.7% accuracy across five activity types.
• Composite energy expenditure model yielded a 10.9% mean absolute percentage error (MAPE).
• MAPE for running, walking, and cycling was 6.6%, 7.9%, and 16.1%, respectively.

Conclusions:

• Integrating thigh-worn accelerometers with ML models accurately classifies physical activity and estimates energy expenditure.
• The novel composite model enhances energy expenditure measurement accuracy.
• This approach supports improved monitoring and assessment of physical activity outside the lab.