•  
  •  
 

USING K-MEANS CLUSTERING TO INDIVIDUALIZE TRAINING FOR COLLEGIATE BASKETBALL ATHLETES FOLLOWING PRE-SEASON PERFORMANCE TESTING

Abstract

Derek A. Crawford1 & Tyler Davis2

1Athlete Health, Performance, & Innovation Laboratory, Department of Nutrition, Kinesiology, and Health, University of Central Missouri, Warrensburg, Missouri; 2Department of Athletics, University of Central Missouri, Warrensburg, Missouri

Player profiling in sports often relies on a mix of tests and coaching decisions that may lead to biases in training plans. K-means clustering offers an unbiased method to categorize players based on performance metrics. This approach can identify strengths and weaknesses in each player, enabling individualized improvement plans. PURPOSE: We used k-means clustering to identify strengths and weaknesses of collegiate men’s basketball athletes based on bio-motor tests. METHODS: Fifteen players completed 12 tests in the pre-season. Tests were grouped based on bio-motor abilities into four categories: speed, agility/change of direction (CoD), and upper and lower extremity strength/power. We determined the optimal number of clusters based on within-cluster sum of squares (WCSS) and principal component analyses identified which features influenced cluster formation. General linear models confirmed which specific performance tests were driving between cluster differences. RESULTS: For speed tests, two clusters were identified (mean WCSS = 16.45) delineated by the linear-curved sprint deficit (mean difference = -3.94 ± 0.82%; t = -4.81, p < .001). Two clusters were identified for agility & CoD (mean WCSS = 12.13) delineated by the change of direction deficit (mean difference = 4.57 ± 0.67%; t = 6.79, p < .001). For lower extremity strength & power, three clusters (mean WCSS = 7.60) were identified primarily delineated by vertical jump with significant differences between all groups (F = 52.2, p < .001). Two clusters were identified for upper extremity strength and power (mean WCSS = 198.72) delineated by the average velocity using 70% of one-repetition maximum during the bench press exercise (mean difference = -0.22 ± 0.06 m/s; t = -3.60, p = .003). CONCLUSIONS: This study highlights k-means clustering as a robust, data-driven method for player profiling. It produced 24 possible unique combinations of player profiles with respect to the tested bio-motor abilities. Within this sample of basketball players, 11 unique profiles were observed across 15 athletes, with no more than two athletes having the same profile. This study demonstrates both the need for highly individualized training plans and outlines a robust method to aid in their implementation. The approach minimizes bias and has applications in sports science.

This document is currently not available here.

Share

COinS