Abstract <p>One of the challenges in determining the physical fitness of cosmonauts is assessing the ventilatory thresholds (VTs), which can be identified through the analysis of gas exchange data collected during cardiopulmonary exercise test. The prediction of VTs often relies on visual inspection of graphical plots of ventilatory equivalents for gas exchange data. The result of V-slope and regression based methods for estimation of ventilatory anaerobic threshold significantly depends on data preprocessing. So, the procedure of VTs estimation can be classified as a poorly formalized problem. That is why, in this paper, we propose a machine learning model based on decision trees to predict VTs. The purpose of the study is to select features by their importance and present an interpretable model for VTs prediction, using minimal data preprocessing. An iterative training of decision trees with different hyperparameters was used to choose significant features from the expert labeled data. The feature importance evaluation showed that, besides basic gas exchange parameters like oxygen uptake, carbon dioxide production, and ventilatory equivalents, metrics derived from indirect calorimetry and measures of lung capacity (e.g., tidal volume, breathing reserve) are also important for VTs prediction. A random forest model was used with the final set of the most relevant features to predict the VTs with an accuracy of approximately 90<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\%\)</EquationSource> <!--BPhysMGU2570311Minkin-m1--> </InlineEquation>. Such a model can be applied to predict VTs for datasets not used in training and to provide useful insights and conclusions based on the data structure. A data clustering approach guided by the accuracy of a machine learning model is proposed.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Machine Learning Model for Ventilatory Thresholds Prediction

  • A. S. Minkin

摘要

Abstract

One of the challenges in determining the physical fitness of cosmonauts is assessing the ventilatory thresholds (VTs), which can be identified through the analysis of gas exchange data collected during cardiopulmonary exercise test. The prediction of VTs often relies on visual inspection of graphical plots of ventilatory equivalents for gas exchange data. The result of V-slope and regression based methods for estimation of ventilatory anaerobic threshold significantly depends on data preprocessing. So, the procedure of VTs estimation can be classified as a poorly formalized problem. That is why, in this paper, we propose a machine learning model based on decision trees to predict VTs. The purpose of the study is to select features by their importance and present an interpretable model for VTs prediction, using minimal data preprocessing. An iterative training of decision trees with different hyperparameters was used to choose significant features from the expert labeled data. The feature importance evaluation showed that, besides basic gas exchange parameters like oxygen uptake, carbon dioxide production, and ventilatory equivalents, metrics derived from indirect calorimetry and measures of lung capacity (e.g., tidal volume, breathing reserve) are also important for VTs prediction. A random forest model was used with the final set of the most relevant features to predict the VTs with an accuracy of approximately 90 \(\%\) . Such a model can be applied to predict VTs for datasets not used in training and to provide useful insights and conclusions based on the data structure. A data clustering approach guided by the accuracy of a machine learning model is proposed.