A Hybrid Unsupervised-Supervised Framework for Player Role Classification in the NBA

Abstract

This study proposes a hybrid analytical framework that combines unsupervised clustering and supervised machine learning methods to systematically reveal the changing patterns of modern NBA player roles. Based on player data from the ten seasons from 2015 to 2025, the study identified seven types: Offensive Orchestrator, Post Anchor, Non-Stretch Center, Roll Big, Midrange Technician, All-Around On-Ball Scorer, and Stretch Wing. This result reshapes the theoretical framework of basketball roles, indicating that the player's role has changed from a static concept of “position” to a dynamic functional category that reflects the characteristics of the “positionless basketball” era. The study used four machine learning models (XGBoost, LightGBM, Random Forest, and Logistic Regression) to verify the reliability of clustering results. Among them, the XGBoost model's accuracy is 84.8%. The results show that, in distinguishing players' roles, technical indicators (such as assist rate, three-point shooting rate, and shooting area distribution) are more indicative than body measurements. Feature importance analysis reveals that offensive autonomy, shooting efficiency, and spatial positioning are key factors in defining modern player roles. These findings provide empirical support for the no-position basketball theory, indicating that the sport is shifting to a professional functional role rather than a traditional position designation.