Estimating NBA players salary share according to their performance on court: A machine learning approach
This work addresses salary prediction for NBA teams and analysts, but it is incremental as it applies an existing machine learning method to a specific sports domain with external validation to avoid overfitting.
The study tackled the problem of predicting NBA players' salary share based on on-court performance by using a Random Forest algorithm to identify key determinants like years of experience and games played, achieving very satisfactory predictions across data from 2017-2019.
It is customary for researchers and practitioners to fit linear models in order to predict NBA player's salary based on the players' performance on court. On the contrary, we focus on the players salary share (with regards to the team payroll) by first selecting the most important determinants or statistics (years of experience in the league, games played, etc.) and then utilise them to predict the player salaries by employing a non linear Random Forest machine learning algorithm. We externally evaluate our salary predictions, thus we avoid the phenomenon of over-fitting observed in most papers. Overall, using data from three distinct periods, 2017-2019 we identify the important factors that achieve very satisfactory salary predictions and we draw useful conclusions.