Data-driven role identification and role-specific performance evaluation in the NBA

The positionless evolution of modern NBA basketball has challenged traditional player classifications, motivating the need for frameworks that identify functional roles and evaluate performance within those roles. However, existing approaches often overlook the richness of play-type frequency data and rarely extend beyond role identification to evaluate role-specific performance. This study proposes a two-step framework to: (i) identify player roles using play-type and advanced metrics; and (ii) assess player performance conditional on these roles. Data from 422 players during the first portion of the 2025–26 NBA regular season were analyzed. Principal component analysis was first applied for dimensionality reduction, followed by k-means clustering to identify distinct player roles based on advanced metrics, shot distribution, and play-type frequencies. In a second step, player performance was evaluated within each role using action-specific efficiency metrics. Five distinct roles emerged: On-ball creators, Rim presences, Spot-up shooters, Off-ball shooters, and Post-up generators. These roles reflect the tactical structure of modern NBA offenses and emphasize the importance of skill-based rather than position-based player classification. Importantly, the proposed framework enables role-aware benchmarking by identifying the most efficient players within each role accounting for their specific on-court responsibilities. By linking role identification with role-specific evaluation, this approach offers a practical and interpretable tool for coaching staffs, analysts, and front offices. It provides insights and role-aware benchmarking for basketball organizations, which may serve as a resource to support in-season coaching decisions, evaluate potential roster fit, or guide player development.