xLength: Predicting Expected Ski Jump Length Shortly after Take-Off Using Deep Learning.

Area of Science:

• Sports Science
• Data Analytics
• Biomechanics

Background:

• Tracking systems are increasingly used in sports, generating vast data for performance analysis.
• Ski jumping generates complex data including 3D trajectory, velocity, and environmental factors.

Purpose of the Study:

• To develop an "expected jump length" (xLength) metric for ski jumping, analogous to xG in soccer.
• To evaluate machine learning models for predicting ski jump length.
• To assess the impact of input data duration on prediction accuracy.

Main Methods:

• Analysis of 2523 ski jumps from 205 athletes across five venues.
• Utilized datasets including 3D trajectory, velocity, ski orientation, wind, and starting gate data.
• Evaluated fully connected neural networks, CNN, LSTM, and ResNet architectures for xLength prediction.

Main Results:

• ResNet models achieved a mean absolute error (MAE) of 5.3 m for new athletes and 5.9 m for new venues.
• Prediction accuracy improved with longer input time series after take-off.
• The developed xLength metric demonstrates potential for real-time broadcasting and expert analysis.

Conclusions:

• Machine learning models, particularly ResNet, can accurately predict ski jump length.
• The xLength metric offers a valuable tool for real-time performance evaluation and post-event analysis in ski jumping.
• This approach enhances sports analytics by quantifying jump performance dynamically.