Performance Prediction and Evaluation in Female Handball Players Using Machine Learning Models

Abstract

Machine learning models are implemented to perform tasks that human beings have difficulty completing. The analysis and prediction of players' performance of specific athletic tasks have increasing importance in both game and training planning. The diversity and complexity of specific types of athletic performance and the mostly nonlinear relationships between them make analysis and prediction tasks complicated when using conventional methods. Therefore, the use of effective machine learning models may contribute to the ability to achieve high accuracy predictions of players' athletic performance. The aim of this study was to evaluate different machine learning models for predicting particular types of athletic performance in female handball players and to determine the significant factors influencing predicted performances by using the superior model. Linear regression, decision tree, support vector regression, radial-basis function neural network, backpropagation neural network and long short-term memory neural network models were implemented to predict the performance of female handball players in countermovement jumps with hands-free and hands-on-hips, 10 meter and 20-meter sprints, a 20-meter shuttle run test and a handball agility specific test. A total of 23 properties and measurements of attributes and 118 instances of training patterns were recorded for each machine learning models. The results showed that the radial-basis function neural network outperformed the other models and was capable of predicting the studied types of athletic performance with R 2 scores between 0.86 and 0.97. Finally, significant factors influencing predicted performance were determined by retraining the superior model. This is one of the first studies using machine learning in sport sciences for handball players, and the results are encouraging for future studies.