ESTABLISHING RULES USING MACHINE LEARNING FOR MODULATING PREDICTIONS OF AFFECT AND EFFORT

Abstract

The aim of this study is to identify through Machine learning rules for predictions modulation of affect (FS) and rating of perceived effort (PSE) related to exercise, analyzing the degree of importance of different biological factors. Each participant (n = 34 [52.9% females], Age = 25.4 ± 6) underwent a test in which they had to predict how much FS and PSE would feel for Sprint before perform it. Subsequently, the respective responses were analyzed using a decision tree algorithm to establish rules for modulation of FS and PSE based on the following characteristics of each subject: fat percentage (%G); muscular mass (%MM); waist and hip ratio (RCQ); self-declared physical activity level (NAF-AUTO); resting heart rate (FCr); sex and age. As result, although not significant, very active individuals predicted an average affective response of 2.28 (95% CI = 0.01 - 4.55; p = 0.06) units higher than those considered insufficiently active. The main rules for participants to predict the lowest FS values ​​were to have the following characteristics, %MM ≤ 48.9 followed by FCr > 73. The highest FS values ​​were %MM > 48.9. The highest PSE values ​​were %G > 16.4, while the lowest values ​​were %G ≤ 16.4. It was concluded that the rules for profile that predicted the higher displeasure was to present less healthy body composition and cardiac capacity, which predicted higher pleasure had opposite characteristics. In the case of PSE, there are two initial profiles divided by %G, and the one with the lowest value predicted less effort, the profile with the highest %G has greater heterogeneity in characteristics.