Heart Rate Variability Based Estimation of Maximal Oxygen Uptake in Athletes Using Supervised Regression Models

Area of Science:

• Sports Medicine
• Physiology
• Machine Learning

Background:

• Wearable heart rate monitors offer physiological insights into athlete well-being and performance.
• Heart rate and heart rate variability (HRV) are physiologically relevant for estimating maximal oxygen uptake (VO2 max).
• Previous data-driven models have utilized heart rate data to estimate cardiorespiratory fitness.

Purpose of the Study:

• To evaluate the utility of HRV features for estimating maximal oxygen uptake (VO2 max) in a large cohort of athletes.
• To compare the performance of different machine learning models in predicting VO2 max using HRV.
• To investigate the impact of feature selection and outlier removal on model accuracy and estimation error.

Main Methods:

• Extracted HRV features from exercise and recovery segments during Graded Exercise Testing for 856 athletes.
• Employed three machine learning models and three feature selection methods to estimate VO2 max.
• Utilized k-Nearest Neighbour for post-modelling outlier removal in training and testing datasets.

Main Results:

• Feature selection increased model accuracy by 5.7% (exercise) and 4.3% (recovery).
• Outlier removal reduced estimation error by 19.3% (exercise) and 18.0% (recovery).
• The 'real-world scenario' model achieved average R values of 0.72 (exercise) and 0.70 (recovery).

Conclusions:

• Heart rate variability is a validated tool for estimating maximal oxygen uptake in large athlete populations.
• Wearable heart rate monitors can effectively contribute to cardiorespiratory fitness assessment in athletes.
• Machine learning approaches combined with HRV analysis offer a promising method for objective athlete monitoring.