PREDICTION OF SPORTS TALENT IN YOUNG THROWERS USING MACHINE LEARNING

Abstract

The objective of this study is to detect any performance factors in athletics throws between 1997 and 2015 in 662 throwers (15.67 ± 1.01 years of the National Program for Sports Technification of the Royal Spanish Athletics Federation using Machine Learning methods by means of algorithms (Logistic Regression, Random Forest and XG Boost). When examining the importance of the variables with reference to performance, the triple jump (0.20) stands out over the rest of the variables: backward overhead shot throw (0.14), arm span (0.11), vertical jump (0.10), body mass (0.20), height (0.07) and flexibility (0.03). In each discipline the triple jump takes the lead in shot put (0.20), discus (0.21) and hammer (0.29) throws, while backward overhead shot throw does in javelin, the variables rearranging themselves in a particular way depending on the discipline. These findings enable the early detection of potential talents as well as their subsequent sport specialization.