Exploring the Application of Pattern Recognition and Machine Learning for Identifying Movement Phenotypes During Deep...

Area of Science:

• Biomechanics
• Kinesiology
• Data Science in Sports

Background:

• Movement screens are vital in sports and rehabilitation for assessing movement competency.
• Current methods often rely on subjective visual analysis of predefined movement features.
• Existing screens may not fully capture whole-body coordination or feature associations.

Purpose of the Study:

• To employ pattern recognition and machine learning to identify whole-body movement phenotypes in functional tasks (deep squat, hurdle step).
• To compare discrete kinematic measures between machine learning-identified movement groups.
• To explore objective classification of movement behaviors.

Main Methods:

• Applied Principal Component Analysis (PCA) to 3D kinematic data from deep squat and hurdle step tasks.
• Utilized Gaussian Mixture Model (GMM) for unsupervised clustering of PCA scores to identify movement phenotypes.
• Employed one-way ANOVA to analyze differences in kinematic features between identified clusters.

Main Results:

• Identified three movement phenotypes for deep squat and right hurdle step, and four for the left hurdle step.
• Observed significant differences in commonly used discrete kinematic measures between these emergent groups.
• Demonstrated significant main effects (p < 0.05) for most discrete kinematic measures across phenotypes.

Conclusions:

• Whole-body movement analysis using pattern recognition and machine learning can objectively define movement phenotypes.
• This approach bypasses the need for *a priori* defined movement features.
• Highlights the importance of considering factors influencing machine learning outcomes in movement analysis.