한국스포츠경영전략연구원

ssjeon@k-spo.co.kr

This study aims to systematically analyze the mechanisms underlying salary determination for NBA players and to develop a data-driven, theoretically grounded compensation framework. Drawing on comprehensive NBA data set spanning from the 1999-2000 to the 2024-25 seasons, the study conducted structured data preprocessing using Python and applied seven machine learning algorithms— Light-GBM, Gradient Boosting, Random Forest, Ridge, Lasso, ElasticNet—were applied to forecast player salaries for the 2017-18 through 2024-25 seasons. Among these, the ElasticNet and GradientBoosting algorithm exhibited the highest explanatory capacity, achieving a coefficient of determination (R²) of 0.755, which accounted for 68.5% of the salary variance for the 2024-25 season. Grounded in integrated theoretical perspectives spanning human capital theory, performance-based compensation theory, and labor market theory, the study meticulously constructed its variable sets to achieve both analytical rigor and theoretical interpretability. This foundational approach enabled the investigation not only of direct performance metrics but also multifaceted contextual and trajectory-based factors influencing remuneration. A distinctive contribution of the analysis lies in its visualization of the discrepancies between performance-focused and multifactorial salary predictions, thereby empirically pinpointing overvalued and undervalued players within the existing compensation system. These visual and statistical insights foster a more nuanced understanding of structural salary inequities in professional basketball. 

Keywords：Artificial Intelligence (AI), Machine Learning, Compensation Prediction, Player Salary Estimation, NBA