Title: Performance Profiles in Youth Basketball Across Different Score Contexts: An Unsupervised Machine Learning Analysis
Author: Dimitrios Pantazis, Christos Kokkotis, Athanasios Chatzinikolaou, et al.
Institution: Department of Physical Education and Sport Science and Department of Occupational Therapy, Democritus University of Thrace, Greece; Journal of Functional Morphology and Kinesiology, 11(3), 282
Publication Date: July 21, 2026
The Problem: Load Data and Box Scores Live in Separate Reports
Teams now collect wearable load data and box scores side by side, but usually read them separately. The load report says how hard players worked; the stat sheet says how well they played. Few studies have asked whether those two signals combine into recognizable patterns, and almost none at the youth level, where physical maturity, training exposure and tactical discipline vary widely.
Context complicates it further: a quarter decided by two points may demand something different from one decided by twelve. This paper asks whether U16 quarters naturally sort into distinct physical plus technical performance profiles.
Methodology: Four Pieces, One Data-Driven Sort
Researchers at Democritus University of Thrace tracked 19 Final 8 games of the Greek U16 Championship, producing 117 team quarter observations split at the median 5 point margin into close (62) and lopsided (55) quarters.
Wearable sensors (the court-demand log): Hip mounted KINEXON units captured distance by speed zone, accelerations, mechanical load and jump load during live play only.
Box score plus advanced metrics (the scouting sheet): Points, assists, steals, eFG%, offensive rating, PIR and possessions were coded by one analyst with near-perfect re-test agreement (kappa 0.91 to 0.95).
K-means clustering (sorting film without labels): Every standardized variable went in, and the algorithm grouped similar quarters with no win or loss tag. Four validation indices pointed to two groups in each context.
Stability check (re-running the tape): 100 restarts produced near identical groupings (ARI 0.965 and 0.948).
Why it Matters: Effort and Output Travel Together, Wins Don't Follow Automatically
Lopsided quarters split on both engines: The higher cluster posted a PIR of 29.7 vs 10.5 (d = 2.04), 5.8 vs 2.6 assists and 21.7 vs 13.1 points while covering 93.5 vs 80.2 meters per minute.
Close quarters split on volume, not efficiency: Total distance (8,411 vs 6,598 m) and mechanical load drove separation. eFG% barely moved (39% vs 37%), but possessions and free throw trips rose.
Running more is not winning more: High clusters won 70.0% vs 45.7% and 56.7% vs 40.6% of quarters, but neither gap was statistically significant (p = 0.145 and 0.313).
Read it as exploratory: One tournament, quarters nested within the same teams, and clusters described by the variables that built them. Close quarter separation was weak (silhouette 0.108).
ACTIONABLE TAKEAWAYS
Pair every load report with the box score: A high distance quarter is only good news if offensive rating and assists rose with it, so review both by quarter.
Use tempo in close games to buy possessions: The high load profile in tight quarters created extra trips and free throws, not better shooting, so pair pace with shot quality work.
Benchmark youth against youth: These U16 players showed higher acceleration load but fewer jumps and less distance per minute than older players, so adult targets do not transfer.