Paper Title
Support Vector Machines With Individual Combinations of Relief-F and mRMR for Predicting VO2MAX From Maximal and Questionnaire Data

Abstract
Maximum oxygen uptake (VO2max) is widely accepted as being a reliable and valid measure of cardiorespiratory fitness (CRF). In this paper, Support Vector Machine (SVM) with individual combinations of Relief-F and minimum redundancy maximum relevance (mRMR) feature selection algorithms has been used to build new VO2max prediction models from maximal and questionnaire data, with the aim to compare the performance of Relief-F with mRMR and to identify the discriminative predictors of VO2max. The dataset is made up of 440 (230 females, 210 males) volunteers ranging in age from 20 to 79, and includes the physiological variables gender, age, body mass (BM) and height; the maximal variables heart rate (HRmax), rating of perceived exertion (RPE), respiratory exchange ratio (RER) and exercise time; and finally the questionnaire variable activity code (AC). The dataset has been randomly divided into training and test sets by applying 10-fold cross validation. The root mean square errors (RMSE's) have been calculated to assess the performance of the prediction models. The results reveal that the prediction model containing all physiological, maximal and questionnaire variables yields the lowest RMSE with 3.79 mL kg-1 min-1. As for the importance of the variables for VO2max prediction; it turns out that exercise time is the most important predictor variable, independent of whether Relief-F or mRMR has been applied on the dataset. Furthermore, it is seen that mRMR in average exhibits slightly better performance than Relief-F for prediction of VO2max. Keywords� Support Vector Machine, Relief-F, mRMR, Maximal Oxygen Uptake, Prediction.