Cricket fast bowling detection in a training setting using an inertial measurement unit and machine learning

Area of Science:

• Sports Science
• Biomechanics
• Machine Learning

Background:

• Fast bowlers face high overuse injury risks, influenced by bowling frequency.
• Inertial Measurement Units (IMUs) can classify sports movements but commercial options are often costly.
• Smartphone accessibility to IMUs presents an opportunity for affordable workload monitoring.

Purpose of the Study:

• To develop and validate machine learning models for automated workload monitoring in elite fast bowlers.
• To assess the impact of data sampling frequency and bowling phase inclusion on model accuracy.
• To enhance the accessibility of workload monitoring for athletes across various IMU specifications.

Main Methods:

• Trained and validated five machine learning models using data from 17 elite fast bowlers.
• Compared models trained on all bowling phases versus only the delivery phase at 250 Hz.
• Evaluated model accuracy with data down-sampled to 125 Hz, 50 Hz, and 25 Hz.

Main Results:

• Models trained solely on the delivery phase achieved high accuracy (>95%), comparable to models using all phases.
• Accuracy remained consistently high (>96%) across all models and down-sampled frequencies.
• Machine learning models demonstrated robustness across different IMU specifications and data rates.

Conclusions:

• Automated workload monitoring for fast bowlers is feasible using machine learning and IMUs.
• Focusing on the delivery phase and utilizing lower sampling rates maintains high accuracy.
• This approach increases the accessibility of crucial workload monitoring for a wider range of athletes.