PREDICTION OF THE PERFORMANCE INDEX RATING OF BASKETBALL PLAYERS IN THE EUROLEAGUE

Abstract

This paper presents and evaluates machine learning (ML) as well as simple custom made approaches for the prediction of the Performance Index Rating of basketball players in the Euroleague. The approaches rely on the large amounts of data, available on the official website of the Euroleague. The data are processed and prepared in order to create a training and evaluation datasets for the application of ML algorithms. The algorithms which were considered are: Naive Bayes, SVM (Support Vector Machines) and KNN (K nearest neighbours). Parameters of each of the algorithms were optimized using a validation set. Besides the ML models, two simple custom made algorithms were designed and implemented by the author of this paper. The best accuracy of 92% was achieved by the SVM algorithm, while the Naive Bayes and KNN algorithms had a lower accuracy of 87% and 72% respectively. The accuracy of the custom made algorithms was significantly lower when compared to the machine learning models.