Multisensory assessment and machine learning for athlete classification in talent identification.

Area of Science:

• Sports Science
• Biomechanics
• Neuroscience

Background:

• Elite sport talent identification faces challenges with maturation and limited objective tools.
• This study investigates visual-vestibular-somatosensory and autonomic (VVS-A) measures for distinguishing elite divers.

Purpose of the Study:

• Identify VVS-A features differentiating podium-level divers from beginners using statistical analysis.
• Evaluate machine learning models for classifying elite diving athletes.
• Analyze classification probabilities using lift-curve analysis.

Main Methods:

• Cross-sectional study employing machine learning classification.
• Sixty participants from an Olympic diving program underwent VVS-A assessments.
• Somatosensory (ankle proprioception), visual, vestibular, and autonomic functions were measured; machine learning models were trained and validated.

Main Results:

• Podium-level divers exhibited superior ankle proprioception and visual-vestibular smooth pursuit.
• No significant group differences were found in voluntary saccades or autonomic metrics.
• A Ridge Logistic Regression model achieved 94.4% accuracy in classifying podium-level athletes (AUC=0.889).

Conclusions:

• Specific VVS-A measures correlate with current performance levels in Olympic diving.
• The study's cross-sectional design and sample limitations prevent definitive conclusions on predictive validity.
• Longitudinal, sport-specific validation is required before applying these findings to talent identification.