Predicting cardiorespiratory parameters based on endurance testing

CUNDRIČ, LARSEN (author), Bosnić, Zoran (mentor), Popović, Dejana (comentor)

Maximal heart rate (HRmax) and Maximal oxygen consumption (VO2max) are measures of adequate effort during an exercise test. Current equations for HRmax and VO2max prediction are not sufficiently accurate. Our aim was to improve those predictions using machine learning (ML). We used a sample of the Fitness Registry and the Importance of Exercise: An International Data Base (FRIEND Registry) with 17,325 healthy individuals (81% males) who performed a maximal cardiopulmonary exercise test (CPX). Mean age was 45.81±12.54 years, HRmax was 162.49±20.07 bpm and VO2max 32.51±11.13 mlO2 kg^-1 min^-1. Different ML algorithms were used to predict HRmax and VO2max: lasso regression, random forests, neural networks, and support vector machine. Prediction accuracy was measured using the root mean squared error (RMSE), relative RMSE (RRMSE), Pearson correlation test and Bland-Altman analysis. The best predictive model was explained using the Shapley’s Additive Explanations (SHAP). For ML predictions we used age, resting heart rate, weight, height and resting systolic and diastolic blood pressure based on RReliefF feature selection. All models improved HRmax and VO2max prediction and decreased RMSE and RRMSE compared to baseline formulas.

• Language: English
• Keywords: machine learning, maximal heart rate, feature selection, maximal oxygen uptake, cardiorespiratory fitness
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2021
• PID: 20.500.12556/RUL-132134
• COBISS.SI-ID: 82279939
• Publication date in RUL: 13.10.2021

Secondary language

• Language: Slovenian
• Title: Napovedovanje kardiorespiratornih parametrov na osnovi testiranja vzdržljivosti

Abstract

Maksimalna srčna frekvenca (HRmax) in maksimalna absorpcija kisika v krvi (VO2max) sta merili napora med testom vzdržljivosti. Obstoječi modeli za napovedi spremenljivk niso dovolj natančne, zato je naš cilj izboljšati napovedi s pomočjo strojnega učenja (ML). Za ta namen smo uporabili vzorec podatkov registra FRIEND (Fitness Registry and the Importance of Exercise: An International Data Base) s 17,325 zdravimi posamezniki (81 % moških), ki so opravili test maksimalne srčno-pljučne vadbe (CPX). Povprečna starost je bila 45,81 ± 12,54 let, HRmax 162,49 ± 20,07 utripov/min in VO2max 32,51 ± 11,13 mlO2 k ^- 1 min ^- 1. Za napovedi smo uporabili različne algoritme ML: regresijo Lasso, naključne gozdove, nevronske mreže in metodo podpornih vektorjev. Natančnost napovedi smo izmerili z uporabo korenjene srednje kvadratne napake (RMSE), relativne RMSE (RRMSE), Pearsonovega korelacijskega koeficienta in analize Bland-Altman. Najboljši model smo razložili z uporabo algoritma SHAP. Kot najpomembnejše vhodne spremenljivke, ki so bile izbrane z uporabo algoritma RReliefF, smo uporabili starost, srčni utrip v mirovanju, težo, višino in sistolični in diastolični krvni tlak v mirovanju. Vsi modeli so izboljšali napoved spremenljivk ter zmanjšali RMSE in RRMSE v primerjavi z dosedanjimi tradicionalnimi modeli.

• Keywords: strojno učenje, maksimalna srčna frekvenca, izbira atributov, maksimalna absorpcija kisika, karodiorespiratorna pripravljenost