The limited availability of calorimetry systems for estimating human energy expenditure (EE) while conducting exercise has prompted the development of wearable sensors utilizing readily accessible methods. We designed an energy expenditure estimation method which considers the energy consumed during the exercise, as well as the excess post-exercise oxygen consumption (EPOC) using machine learning algorithms. Thirty-two healthy adults (mean age = 28.2 years; 11 females) participated in 20 min of aerobic exercise sessions (low intensity = 40% of maximal oxygen uptake [VO2 max], high intensity = 70% of VO2 max). The physical characteristics, exercise intensity, and the heart rate data monitored from the beginning of the exercise sessions to where the participants' metabolic rate returned to an idle state were used in the EE estimation models. Our proposed estimation shows up to 0.976 correlation between estimated energy expenditure and ground truth (root mean square error: 0.624 kcal/min). In conclusion, our study introduces a highly accurate method for estimating human energy expenditure during exercise using wearable sensors and machine learning. The achieved correlation up to 0.976 with ground truth values underscores its potential for widespread use in fitness, healthcare, and sports performance monitoring.
Keywords: energy expenditure; excess post-exercise oxygen consumption; heart rate; machine learning.