The informative power of heart rate along with machine learning regression models to predict maximal oxygen consumption and maximal workload capacity

Abstract

Prediction of maximal oxygen consumption (VO_2max) and maximal workload capacity (MWC) through submaximal exercise tests is an important topic for sports sciences. Numerous studies highlighted the predictive power of submaximal heart rate (HR) and oxygen consumption (VO₂) in predicting VO_2max and MWC. The challenge is achieving the best possible precision and accuracy by identifying the best predictors and regression models. This project assessed the performance of different indexes along with machine learning regression models to estimate VO_2max and MWC. Predictors consisted of biodata (age, weight, and height) along with different combinations of change-scores of HR and VO₂ between 0–50 Watts, 50–65 Watts, and 65–80 Watts (Δ0–50, Δ50–65, and Δ65–80, respectively). The use of biodata + HR Δ50–65 + HR Δ65-80 via a Squared Exponential Gaussian Process Regression model resulted in the best performance in predicting VO_2max, while the use of biodata + HR Δ0–50 via a Robust Linear Regression model resulted in the best performance in predicting MWC. These results suggest that information provided by HR only during submaximal exercise offers the best predictive mean for estimating VO_2max and MWC, while the use of VO₂ changes or its addition along with HR changes does not improve predictions. Moreover, different predictors need to be selected for the best estimation of VO_2max and MWC. Change-scores refer to absolute value changes, providing information to develop athlete assessment protocols through standardized workloads. These results show practical applicability for sports assessments to be performed indirectly, rapidly, sub-maximally, and through the simple measurement of HR.