Predicting athletic performance from physiological parameters using machine learning: Example of bocce ball

Abstract

Machine learning (ML) is an emerging topic in Sports Science. Some pioneering studies have applied machine learning to prevent injuries, to predict star players, and to analyze athletic performance. The limited number of studies in the literature focused on predicting athletic performance have adopted the cluster-then-predict classification approach. However, these studies have used the independent variable to represent athletic performance at both the clustering and classification stages. In this study we used only physiological parameters in the classification of bocce athletes. Their performance classes were predicted with high accuracy, thus contributing new findings to the literature. The support vector machines-radial basis function (SVM-RBF) kernel correctly predicted all athletes from the high-performance bocce player (HPBP) cluster and 75% of the athletes in the low-performance bocce player (LPBP) cluster. Using machine learning to predict athletic performance from balance data was found to be a time-saving approach for selecting high-potential bocce athletes.