Performance Profiles in Youth Basketball Across Different Score Contexts: An Unsupervised Machine Learning Analysis

Area of Science:

• Sports Science
• Data Science

Background:

• Advanced analytics and machine learning are increasingly used in basketball performance analysis.
• Understanding factors influencing offensive efficiency and match dynamics is crucial.

Purpose of the Study:

• To identify basketball performance profiles using unsupervised machine learning.
• To examine physical load and performance indicators differentiating these profiles.

Main Methods:

• K-means clustering was applied to stratified team-quarter observations (large vs. small score differences).
• Multiple indices (Elbow, Silhouette, Calinski-Harabasz, Davies-Bouldin) were used for cluster evaluation.
• Cluster stability was assessed using Adjusted Rand Index (ARI), and visualization employed t-distributed stochastic neighbor embedding (t-SNE).

Main Results:

• Significant differences were found in external load variables (jump load, distance covered, acceleration load) between clusters (pFDR < 0.001).
• Higher movement intensity clusters showed superior basketball performance and offensive efficiency indicators.
• Cluster membership was not significantly associated with quarter outcomes, despite higher winning proportions in higher-performance clusters.

Conclusions:

• Basketball performance profiles are primarily distinguished by external load characteristics, especially movement intensity.
• Offensive performance indicators also play a key role in differentiating player profiles.
• Integrating both physical and technical performance indicators is essential for effective basketball performance analysis.