Developing new VO2max prediction models from maximal, submaximal and questionnaire variables using support vector machines combined with feature selection.

Maximal oxygen uptake (VO2max) is an essential part of health and physical fitness, and refers to the highest rate of oxygen consumption an individual can attain during exhaustive exercise. In this study, for the first time in the literature, we combine the triple of maximal, submaximal and questionnaire variables to propose new VO2max prediction models using Support Vector Machines (SVM's) combined with the Relief-F feature selector to predict and reveal the distinct predictors of VO2max. For comparison purposes, hybrid models based on double combinations of maximal, submaximal and questionnaire variables have also been developed. By utilizing 10-fold cross-validation, the performance of the models has been calculated using multiple correlation coefficient (R) and root mean square error (RMSE). The results show that the best values of R and RMSE, with 0.94 and 2.92mLkg-1min-1 respectively, have been obtained by combining the triple of relevantly identified maximal, submaximal and questionnaire variables. Compared with the results of the rest of hybrid models in this study and the other prediction models in literature, the reported values of R and RMSE have been found to be considerably more accurate. The predictor variables gender, age, maximal heart rate (MX-HR), submaximal ending speed (SM-ES) of the treadmill and Perceived Functional Ability (Q-PFA) questionnaire have been found to be the most relevant variables in predicting VO2max. The results have also been compared with that of Multilayer Perceptron (MLP) and Tree Boost (TB), and it is seen that SVM significantly outperforms other regression methods for prediction of VO2max.