Determination of the athletes' anaerobic threshold using machine learning methods. (March 2022)

• Record Type: Journal Article
• Authors: Chikov, Alexander; Egorov, Nikolay; Medvedev, Dmitry; Chikova, Svetlana; Pavlov, Evgeniy; Drobintsev, Pavel; Krasichkov, Alexander; Kaplun, Dmitry
• Is Part Of: Biomedical signal processing and control. Volume 73(2022)
• Publication Date: 2022-03
• Subjects: Anaerobic threshold -- Cardiopulmonary exercising test -- Machine learning -- Model interpretability -- LIME; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103414
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Defining an anaerobic threshold (AT) is challenging at the expert level and is characterized by high variability among experts. Cardiopulmonary exercising test was used to get physiological indicators at the AT. The model was trained on 1273 observations using 4 machine learning methods: Linear Regression, Random Forest Regression, Gradient Boosting, and Support Vector Regression (SVR). The maximum accuracy was achieved in the model based on the SVR algorithm. The possibility of applying the explanatory algorithm LIME to the developed model for interpreting the obtained values and identifying individual patterns limiting physical performance is shown. Abstract: Physiological indicators at the anaerobic threshold (AT) are an important diagnostic criterion for determining the level of an athlete's fitness and one of the starting points for planning and adjusting the training process. To develop the model for determining athletes' AT, the results of 1273 observations of athletes aged 18–35 years were processed. Athletes performed a stepwise cardiopulmonary exercising test (CPET) on the treadmill to failure. Linear Regression, Random Forest Regression, Gradient Boosting, and Support Vector Regression (SVR) from the Scikit-learn library were used to determine the physiological parameters of energy supply at the AT. The best quality metrics for determining the AT were obtained by SVR, where the coefficient of determination for heart rate (HR), respiratory minute volume (V'E), oxygen consumption (VO2), emission of carbon dioxide (VCO2), oxygen pulse (O2/HR) was 0.82, 0.90, 0.87, 0.86, and 0.91, respectively. The special significance of the obtained model lies in the fact that it can be used to identify indicators and their quantitative values that limit the further development of the AT-based technique to plan and correct a training process. This feature was provided by the Local Interpretable Model-agnostic Explanations (LIME). LIME was used to explain the prediction of the developed model in an interpreted and accurate way by studying the model locally around the prediction. The developed model for determining the AT opens up new opportunities in the interpretation of CPET, will allow researchers to identify individual patterns that affect the test result, and, consequently, give more accurate recommendations for correcting the athletes' training process.