A Machine-learning Approach for the Prediction of the Anaerobic Mechanical Power Outputs from Submaximal Exercise Stress Test

During submaximal graded exercise stress test, the active muscles utilize mainly the anaerobic metabolism toward the very end of the test. Therefore, it is hypothesized that submaximal exercise stress test includes hidden information of anaerobic reservoirs and can potentially be translated into power outputs.

Aim: To develop a new computational simulative predictive model, enable to predict the anaerobic capacity from a submaximal graded exercise stress test.

Methods: We collected data from cardiopulmonary exercise stress test (CPET) and from the Wingate test (WAnT) from 88 individuals. We have developed a predictive model which choses the best aerobic feature at each iteration in order to predict the outputs of the WAnT.

Results: We were able to receive equations which can highly predict the peak power and mean power with r = 0.92 and 0.9, and RMSE equal to 79 and 69, respectively. The predictive equation for peak power [w] was:

Peak powe [w] = 644 + 77 * (gender) + 140 * (calculated VO2max) - 90 * (VO2 at the anaerobic thershold).

Conclusions: Using such a model reduces the maximal intensity requires to achieve the anaerobic outputs, using both WAnT and maximal aerobic stress test.