Constructing energy expenditure regression model using heart rate with reduced training time.
Conference Paper
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
Accurate estimation of energy expenditure (EE) is a key enabler for many applications of healthcare and wellness. Heart rate (HR) based EE estimation methods typically require extensive training time to establish a relationship between HR and EE. In this work, we propose a method where just the few most representative EE-HR data pairs are used to train the estimation model. Furthermore, we present a systematical methodology based on the ranking of the correlation coefficients between EE and HR to find the least amount of EE-HR data pairs required for training while satisfying the constraint of estimation accuracy. During the experimental evaluation, while the study participants walk and run on a treadmill, our method is compared to three different training paradigms: training the EE-HR model 1) using all available data collected during the experiment, 2) using the EE-HR data only during speed changes (or during monotonic HR changes) and 3) using the EE-HR data pairs collected during constant speed. The results show that our method could maintain a comparable EE estimation performance as shown by only 2~4% changes on the coefficient of variation of root-mean-squared error (CV(RMSE)) for the testing dataset while saving nearly 91-97% training time for each individual.
name of conference
2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)