Application of Hierarchical Clustering to NBA Player Statistics (2024–2025 Season)

Abstract

Hierarchical clustering was applied to a dataset containing aggregated statistics for 731 NBA players from the 2024–2025 season. The goal was to uncover performance profiles without predefined labels, using an unsupervised, distance-based approach. We considered quantitative variables capturing offensive output (points, rebounds, assists), usage/volume (minutes played, field-goal and free-throw attempts), and defensive contributions (steals, blocks), together with indicators such as turnovers and personal fouls. After preprocessing to make variables comparable across scales (normalization), a dissimilarity matrix was computed and a dendrogram was constructed; the final partition was selected by cutting the hierarchy using separation and internal consistency criteria. The analysis yielded three clearly differentiated groups: a high-usage, high-production cluster; a role/support cluster with lower usage and more balanced contributions; and an intermediate cluster characterized by moderate involvement paired with relatively efficient production. Overall, the resulting segmentation supports player comparison, summarizes dominant season patterns, and provides a practical baseline for downstream tasks such as roster construction, and decision support.