Beyond the box score: Clustering offensive play types to reveal game styles in professional basketball

This study examined offensive tactical patterns in professional basketball and their relationships with scoring performance under varying contextual factors. Data from 982 games in the 2023-2024 Chinese Basketball Association regular season were analyzed. A k-means clustering procedure identified three offensive archetypes: Individual Creation Style (ICS), Pick-and-Roll Variation Style (PNRS), and Screen-Post Synergy Style (SPSS); random forest classifiers with Shapley Additive exPlanations confirmed cluster separations. Correlation analyses showed that isolations and pick-and-roll (PnR) handler plays in ICS were moderately associated with two-point shots made (r = 0.31) and accuracy (r = 0.30); hand-off actions in PNRS weakly correlated with two-point shot accuracy (r = 0.12); and off-ball screens in SPSS were moderately related to three-point shots made (r = 0.31). Poisson regression models revealed contextual effects. Specifically, PnR handler plays increased against stronger opponents (p < 0.001), isolations occurred more frequently in faster-paced games (p = 0.002); hand-off completions were greater in wins (p = 0.034) and against stronger than weaker opponents (p = 0.034); and off-ball screens increased against medium-level opponents (p = 0.011) and after longer travel distances (p = 0.033). These findings provide a scalable analytical framework connecting tactical structures, scoring outcomes, and situational constraints in professional basketball.