An Ensemble of Long Short-Term Memory Models to Automatically Detect End-Range Movement Patterns in Men's...

Area of Science:

• Sports Science
• Biomechanics
• Machine Learning in Sports

Background:

• Evaluating end-range movements in tennis is crucial for quantifying high-intensity load exposure and player physical capabilities.
• Current methods for identifying these movements are labor-intensive and lack efficiency.

Purpose of the Study:

• To develop and evaluate an automated system for detecting end-range movements in professional male tennis players.
• To leverage three-dimensional pose model data and machine learning for enhanced sports performance analysis.

Main Methods:

• Utilized three-dimensional pose model data from male competitors at the 2024 Australian Open.
• Employed an ensemble of 10 long short-term memory (LSTM) models to classify coach-identified end-range movements.
• Applied an average prediction value with a class prediction threshold of 0.63 for optimal performance.

Main Results:

• The best-performing LSTM ensemble achieved an F1-score of 0.944.
• Demonstrated high accuracy (95.9%), precision (97.8%), and recall (91.2%) in classifying end-range movements.
• Successfully validated a novel, practical method for automatic detection of demanding tennis movements.

Conclusions:

• An automated pipeline using pose data and machine learning can effectively quantify high-intensity movement exposures in professional tennis.
• This approach enhances post-match analysis, providing descriptive statistics for load monitoring and player management.
• Offers a significant advancement for sports scientists and coaches in understanding player physical demands.