PREDICTING PHYSICAL PERFORMANCE OF COSMONAUTS: PIONEERING EXPERIMENTS ON SELECTION OF A MATHEMATICAL MODEL

Changing from orbital flights to exploration of remote space will require development of an automated onboard system of monitoring, optimization and predicting cosmonauts' physical performance. The paper proposes a method of predicting the physiological cost of work using the data about daily training with the help of machine learning methods. Mathematical models were constructed for 3 cosmonauts assigned for 6-month missions. Prediction was made using the linear regression, decision trees and random wood. Model quality was assessed and retraining confirmed using R2 (proportion of the explained dispersion). The highest value of metrics R2 = 0.61 was calculated for the random wood. Distance covered in the treadmill passive mode was found to make the largest contribution to the predicted value.