Analysis of Playing Styles of NBA Players Using the K-Medoids Method
Abstract
The evolution of basketball in the NBA has shifted from traditional five-position roles to more flexible, skill-based playstyles. This study explores the classification of NBA players based on their individual performance metrics using the K-Medoids clustering method. Data from the 2015–2016 to 2024–2025 NBA seasons were collected and processed using the CRISP-DM framework. After data standardization and dimensionality reduction with PCA, the K-Medoids algorithm was applied to group players into distinct clusters. Evaluation using Davies-Bouldin Index (DBI) and Silhouette Score confirmed that a three-cluster configuration yielded the best cohesion and separation. The identified clusters reflect distinct roles such as elite scorers, defensive big men, and versatile contributors, providing valuable insights for team composition and strategy optimization.
References
Phadke V, Pai O. The Rise of Neo-Positions in Basketball [Internet]. 2021. Available from: https://www.bruinsportsanalytics.com/post/neo_positions
Levy I. How the Warriors evolved small ball and, in the process, the NBA [Internet]. 2015. Available from: https://www.si.com/the-cauldron/2015/10/12/golden-state-warriors-nba-title-small-ball-steph-curry
Gaurav. NBA records broken during the 2015-16 season [Internet]. 2019. Available from: https://www.essentiallysports.com/nba-records-broken-during-the-2015-16-season/
Pilsbury D. Evaluating Combinations of Play Styles in the NBA by. 2023;
Berkhin P. A survey of clustering data mining techniques BT - Grouping Multidimensional Data. Group Multidimens Data [Internet]. 2006;(c):25–71. Available from: http://dx.doi.org/10.1007/3-540-28349-8_2%5Cnpapers2://publication/doi/10.1007/3-540-28349-8_2
Wirth R, Hipp J. CRISP-DM: towards a standard process model for data mining. Proceedings of the Fourth International Conference on the Practical Application of Knowledge Discovery and Data Mining, 29-39. Proc Fourth Int Conf Pract Appl Knowl Discov Data Min [Internet]. 2000;(24959):29–39.
