Machine learning para a predição da carga de treinamento de um atleta da Seleção Brasileira de Taekwondo

Abstract

The aim of the present study was to adjust a predictive model for the training load of an athlete from the Brazilian Taekwondo team. 122 training sessions were analyzed and the Wellness_Being scale, countermovement jump height mean (CMJmean), and mean heart rate (HR_MED) were used as dependent variables. To predict the rate perception of exertion at three levels, Easy, Moderate and High (RPE_class), the classification Decision Tree, Random Forest, Bagging and Boosting models were adjusted. After carrying out the tests, the Decision Tree showed an accuracy of 26.09%, the Random Forest showed an accuracy of 17.39%, the Bagging method showed an accuracy of 56.52% and Boosting showed an accuracy of 65.21%. In conclusion, the Bossting method performed better and was more suitable for adjusting the decision tree model to predict the RPE_class of the training, but should be used with caution.