Hierarchical Linear Models for Energy Prediction using Inertial Sensors: A Comparative Study for Treadmill Walking.

Area of Science:

• Biomechanics
• Wearable technology
• Exercise physiology

Background:

• Accurate tracking of calories expended during walking is crucial for health and lifestyle interventions.
• Inertial sensors offer a promising method for measuring physical activity, but normalization across individuals remains a challenge.
• Existing methods like weight scaling require population-specific validation.

Purpose of the Study:

• To evaluate an inertial sensor-based hierarchical model for estimating energy expenditure during walking across a diverse population.
• To identify optimal movement and physiological features for accurate energy expenditure prediction.
• To compare the hierarchical model's performance against subject-specific and weight-scaled regression models.

Main Methods:

• Utilized inertial sensor data to develop and test a hierarchical modeling approach for energy expenditure estimation.
• Determined optimal features, finding periodicity-based features more accurate for cross-population generalization.
• Compared hierarchical models with subject-specific regression and weight exponent scaled models.

Main Results:

• Periodicity-based features demonstrated higher accuracy (p<0.1) for generalizing across populations.
• Weight was identified as the most accurate physiological parameter (p<0.1) for prediction.
• Subject-specific models outperformed weight-scaled models, while the hierarchical model initially performed worse.

Conclusions:

• Hierarchical modeling, particularly with informed priors, shows potential for generalized energy expenditure prediction in clinical settings.
• The approach can achieve prediction errors comparable to subject-specific models with extensive training data.
• This technique offers a promising avenue for personalized health monitoring and intervention.