Successful sports performance depends on several physiological and physical fitness components. It is essential to know which fitness features are most important for performance. Sports fitness components are often multicollinear, and the relationship is complex. So there is a need to use more sophisticated methods that can deal with complex multicollinear data. Hence machine learning algorithms are used along with conventional statistical methods for important physiological and physical fitness feature selection in the archery performance of Indian archers. Recursive feature elimination and Boruta algorithm using random forest along with conventional statistical methods are used for feature selection. The root mean square error of stepwise regression was 21.262, recursive feature elimination with 15 features was 19.450 and random forest with 15 features was 8.417 in the training dataset. Further, the root mean square error for random forest with eight confirmed important features was 9.003 and 8.716 for ten non-rejected features in the training dataset. Out of fifteen features, eight features confirmed important are maximum bow hold time, sub-maximal oxygen intake, peak power, average power, core abdominal strength, age, weight, and body fat, while acceleration speed and maximum oxygen intake are tentatively important. Machine learning Boruta algorithm using random forest performs better than traditional statistical and recursive feature elimination method for selecting features as well as predicting performance in unseen data. Thus, eight important features identified through Boruta algorithm are useful to develop battery of test, monitor athletes, and alter the training regimens in real-time and talent selection in the archery.
© Copyright 2021 Advances in Signal and Data Processing. Select Proceedings of ICSDP 2019. Veröffentlicht von Springer Singapore. Alle Rechte vorbehalten.