Machine learning-based classification of ice hockey skating tasks using kinematic data.

Area of Science:

• Biomechanics
• Sports Science
• Machine Learning

Background:

• Accurate identification of skating techniques is crucial for ice hockey performance analysis.
• Previous research has explored various methods for analyzing player movements, but automated identification of specific skating tasks remains a challenge.

Purpose of the Study:

• To evaluate the effectiveness of machine learning models in identifying distinct ice hockey skating tasks using body segment kinematic data.
• To compare the performance of different machine learning models and identify which body segments provide the most predictive kinematic data for skating task classification.

Main Methods:

• Four machine learning models (XGBoost, Support Vector Machine, Random Forest) were employed to classify four ice hockey skating tasks.
• Kinematic data, specifically linear accelerations of the center of mass from the trunk, pelvis, thigh, shank, and foot segments, were used as input features.
• A 10-fold cross-validation stratified by participant was utilized for model training and evaluation.

Main Results:

• Machine learning models achieved high accuracy, ranging from 86.5% to 98.9%, in identifying skating tasks.
• The pelvis segment demonstrated the highest predictive performance, followed by the trunk and foot segments.
• The thigh segment generally showed lower accuracy compared to other body segments across all tested models.

Conclusions:

• Body segment kinematic data, particularly from the pelvis, trunk, and foot, can be effectively used with machine learning for automated identification of ice hockey skating tasks.
• The choice of body segment kinematic data significantly influences the prediction performance of machine learning models.
• This study offers valuable insights for advancing sports analytics and player performance assessment in ice hockey through automated movement analysis.