Hierarchical approaches to estimate energy expenditure using phone-based accelerometers

Area of Science:

• Biomedical Engineering
• Exercise Physiology
• Wearable Technology

Background:

• Physical inactivity increases risks for chronic diseases like cancer, heart disease, stroke, and diabetes.
• Walking is an accessible activity to mitigate sedentary behavior.
• Accurate, individualized energy expenditure assessment is needed for tailored interventions.

Purpose of the Study:

• To develop and compare algorithms for predicting energy expenditure during walking.
• To utilize mobile phone accelerometers and incorporate multiple anthropometric features.
• To assess model performance with limited training data and varying speeds.

Main Methods:

• Developed algorithms extending previous work to include arbitrary anthropometric descriptors.
• Tested various models: nearest neighbor, weight-scaled, hierarchical linear, multivariate, and speed-based.
• Used mobile phone accelerometers to measure movement during steady-state treadmill walking.

Main Results:

• Hierarchical linear models incorporating size-based features (weight, height, BMI) showed lower prediction errors.
• Weight was the best individual anthropometric descriptor, followed by height.
• Hierarchical models outperformed others in predicting energy expenditure with limited training data and demonstrated uniform interpolation across speeds.

Conclusions:

• Hierarchical models utilizing anthropometric data and accelerometer measurements offer accurate, individualized energy expenditure prediction for walking.
• These models are robust even with limited training data, enabling effective personalized physical activity interventions.
• The findings support the use of mobile sensing for objective physical activity monitoring and health promotion.