Performance prediction of taekwondo athletes using machine learning

Abstract

Purpose The purpose of this paper was analyzing the performance of elite taekwondo athletes with machine learning approaches through achieving three objectives: (1) clustering the performance of elite taekwondo athletes into four clusters of excellent to poor performance, (2) determining the most effective physical characteristics for performance in this sport, and (3) predicting their medal-winning in the world competitions. Design/methodology/approach Descriptive and predictive models were employed based on the National Olympic Academy dataset of Iranian taekwondo athletes’ physical fitness and anthropometric records which were collected during 1996–2019. Findings In the female (999 records) and male (1560 records) datasets, SOM-kmeans and SOM-spectral clustering algorithms (with average test efficiency of Silhouette as 80%, 79% and Davies-Bouldin as 20%, 34%) were applied and 4 clusters have been obtained based on different physical functions of taekwondo athletes and the number of medals allocated in each cluster. Semi-supervised learning model using the CPLE-Learning algorithm in both female and male datasets demonstrated the possibility of winning medals in the competitions. The accuracy of predicting gold, silver and bronze medals in female dataset were 68%, 59% and 73%, and in the male dataset have been found 58%, 61% and 54%, respectively. Originality/value The results indicate that machine learning algorithms’ ability in sports performance analysis provides valuable information for the management of taekwondo athletes and better planning to prepare them physically.