Research Article
Towards a Generalized Regression Model for On-body Energy Prediction from Treadmill Walking
@INPROCEEDINGS{10.4108/icst.pervasivehealth.2011.246026, author={Harshvardhan Vathsangam and Adar Emken and Todd Schroeder and Donna Spruijt-Metz and Gaurav Sukhatme}, title={Towards a Generalized Regression Model for On-body Energy Prediction from Treadmill Walking}, proceedings={5th International ICST Conference on Pervasive Computing Technologies for Healthcare}, publisher={IEEE}, proceedings_a={PERVASIVEHEALTH}, year={2012}, month={4}, keywords={Index Terms--- Accelerometer Bayesian Linear regression Gyroscope Hierarchical Linear Model}, doi={10.4108/icst.pervasivehealth.2011.246026} }
- Harshvardhan Vathsangam
Adar Emken
Todd Schroeder
Donna Spruijt-Metz
Gaurav Sukhatme
Year: 2012
Towards a Generalized Regression Model for On-body Energy Prediction from Treadmill Walking
PERVASIVEHEALTH
ICST
DOI: 10.4108/icst.pervasivehealth.2011.246026
Abstract
Walking is a commonly available activity to maintain a healthy lifestyle. Accurately measuring calories expended during walking can improve user feedback and intervention measures. Inertial sensors are a promising measurement tool to achieve this. An important aspect in mapping inertial sensor data to energy expenditure is the question of normalizing across physiological parameters. Common approaches such as weight scaling require validation for each new population. An alternative is to use a hierarchical model to model subject-specific parameters at one level and cross-subject parameters connected at a higher level. In this paper, we evaluate an inertial sensor-based hierarchical model to measure energy expenditure across a target population. We determine the optimal physiological parameter set to represent data. Weight is the most accurate parameter (p<0.1) measured as percentage prediction error. We compare the hierarchical model with a subject-specific regression model and weight exponent scaled models. Subject-specific models perform significantly better (p<0.1 per subject) than weight exponent scaled models at all exponent scales whereas the hierarchical model performed worse than both. We study the effect of personalizing hierarchical models with initial conditions for training subject-specific models with limited training data. Using an informed prior produces similar errors to using a subject-specific model with large amounts of training data (p<0.1 per subject).