Beyond Playing Positions: Categorizing Soccer Players Based on Match-Specific Running Performance Using Machine...

Area of Science:

• Sports Science
• Performance Analysis
• Machine Learning in Sports

Background:

• Traditional categorization of soccer players by position may not accurately reflect physical match performance.
• Understanding player-specific running demands is crucial for effective training and performance evaluation.

Purpose of the Study:

• To compare the effectiveness of categorizing soccer players by playing position versus unsupervised machine learning based on running performance.
• To determine which categorization method better identifies distinct player subgroups with similar physical demands.

Main Methods:

• Collected match-specific running data from 40 elite male soccer players over two seasons.
• Utilized k-means clustering based on running performance metrics (total distance, low, moderate, high-intensity running, sprint distance).
• Compared clustering results with traditional playing position categories, analyzing variance and standardized differences between subgroups.

Main Results:

• Clustering based on running performance revealed significantly less variance within groups and larger differences between groups compared to playing positions.
• Distinct clusters showed significant differences in sprint capacity and high-intensity running, which were not evident between playing positions.
• Sprint speed differed between identified sprint and high-intensity endurance clusters.

Conclusions:

• Unsupervised machine learning provides a more robust method for categorizing soccer players based on match-specific running performance.
• This data-driven approach aids in more accurate performance evaluation and personalized physical training program optimization.
• Machine learning-based categorization offers superior insights into player physical profiles compared to traditional positional groupings.