Constructing energy expenditure regression model using heart rate with reduced training time.

Area of Science:

• Biomedical Engineering
• Exercise Physiology
• Wearable Technology

Background:

• Accurate energy expenditure (EE) estimation is crucial for healthcare and wellness applications.
• Traditional heart rate (HR) based EE methods demand extensive training periods.
• Optimizing EE estimation requires efficient model training strategies.

Purpose of the Study:

• To develop a novel method for EE estimation using minimal, representative HR-EE data pairs.
• To establish a systematic approach for identifying the optimal subset of data for training EE models.
• To significantly reduce the computational time required for training personalized EE estimation models.

Main Methods:

• Proposed a training methodology utilizing a small, curated set of EE-HR data pairs.
• Developed a systematic approach based on correlation coefficients to select the minimal required training data.
• Compared the proposed method against three traditional training paradigms using treadmill data (all data, speed changes, constant speed).

Main Results:

• The proposed method achieved comparable EE estimation accuracy to models trained with all data.
• Achieved significant reductions in training time, saving 91-97% per individual.
• Demonstrated minimal changes (2-4%) in the coefficient of variation of root-mean-squared error (CV(RMSE)) on the testing dataset.

Conclusions:

• A highly efficient method for training EE-HR models has been developed.
• This approach drastically cuts down training time without compromising estimation accuracy.
• The findings support the use of optimized data selection for personalized EE estimation in wellness and healthcare.