Classification of Alpine Skiing Styles Using GNSS and Inertial Measurement Units.

Area of Science:

• Sports Science
• Biomechanics
• Data Science

Background:

• Alpine skiing involves distinct turning styles: snowplow, snowplow-steering, drifting, and carving.
• These styles vary in speed, control, and complexity, making visual differentiation possible but data-driven classification challenging.
• Objective, data-driven methods for classifying skiing turns are currently underexplored.

Purpose of the Study:

• To develop and evaluate a data-driven method for classifying alpine skiing turn styles.
• To utilize Global Navigation Satellite System (GNSS) and inertial measurement unit (IMU) data for this classification task.

Main Methods:

• Collected data from 20 advanced/expert skiers performing 2000 turns using IMU sensors on ski boots and a GNSS-equipped mobile phone.
• Extracted and selected relevant features from sensor data using recursive feature elimination.
• Employed and compared three machine learning classifiers: decision trees, random forests, and gradient boosted decision trees.

Main Results:

• Random forests and gradient boosted decision trees achieved high classification accuracies: over 93% for parallel turns (drifting/carving) and 88% for non-parallel turns (snowplow/snowplow-steering).
• Decision trees showed lower accuracy compared to ensemble methods.
• Identified key features crucial for distinguishing between different skiing turn styles.

Conclusions:

• Inertial Measurement Unit (IMU) and GNSS data provide a reliable basis for classifying alpine skiing turn styles.
• Advanced machine learning techniques, particularly random forests and gradient boosted trees, are effective for this classification.
• Future improvements may involve incorporating environmental data such as slope and weather conditions.