Prediction of soccer clubs’ league rankings by machine learning methods: The case of Turkish Super League

Abstract

The aim of this research is to predict league rankings through various machine learning models using technical and physical parameters. This study followed a longitudinal observational analytical design. The SENTIO Sports optical tracking system was used to measure the physical demands and technical practices of the players in all matches. Then, the data regarding the last three seasons of the Turkish Super League (2015–2016, 2016−2017, and 2017−2018), was collected. In this research, league rankings were estimated using three machine learning methods: Artificial Neural Networks (ANN), Radial Basis Function (RBFN), Multiple Linear Regression (MLR) with technical and physical parameters of all seasons. Performances were evaluated through R2, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). Prediction results of the models are the following: ANN Model; R2 = 0.60, RMSE = 3.7855 and MAE = 2.9139, RBFN Model; R2 = 0.26, MAE = 3.6292 and RMSE = 4.5168, MLR Model; R2 = 0.46, MAE = 3.4859 and RMSE = 4.2064. These results showed that ANN can be used as a successful tool to predict league rankings. In the light of this research, coaches and athletic trainers can organize their training in a way that affects the technical and physical parameters to change the results of the competition. Thus, it will be possible for teams to have a better place in the league-end success ranking.