Prediction of the duration of maximal exercise test in professional adolescent football players based on the cardiorespiratory signals – a pilot study

This pilot study aimed to evaluate the potential of machine learning models utilizing parameters estimated from cardiorespiratory signals obtained during rest in predicting the duration of maximal cardiopulmonary exercise tests (CPET) in professional adolescent football players. The study involved a group of 36 male athletes, whose cardiac and respiratory signals were recorded in a supine position for at least 5 minutes. Heart rate variability, statistical features from respiratory signal and features from causal and information domains quantifying the interdependency between cardiac and respiratory signals were calculated and later used for the analysis. The most relevant features (Pearson correlation with CPET duration over 0.2) were used for the machine learning modeling with leave-one-out validation applied. Models demonstrated promising results with a mean absolute percentage error of 17%, mean absolute error of 129 seconds, root mean squared error of 170 seconds, R² score of 0.52, and a Pearson correlation coefficient of 0.74. Explainable artificial intelligence techniques provided insights into the influence of individual features, showing the primary importance of the cardiac parameters but also highlighting the need for incorporating the information from respiratory signal and cardiorespiratory interdependencies.