Supervised Machine Learning Applied to Wearable Sensor Data Can Accurately Classify Functional Fitness Exercises...

Area of Science:

• Biomechanics and Human Movement Analysis
• Machine Learning Applications in Sports Science
• Wearable Sensor Technology

Background:

• Classifying human movements is crucial for applications like sports performance and clinical assessment.
• Challenges in movement classification include data complexity and individual variability.
• Automatic classification of functional fitness drills is needed for effective training and monitoring.

Purpose of the Study:

• To develop and validate a method for automatic classification of four common functional fitness drills.
• To evaluate the performance of supervised learning models using inertial measurement unit data.
• To assess the impact of sensor placement and data segmentation on classification accuracy.

Main Methods:

• Utilized five inertial measurement units (IMUs) on participants' limbs and trunk.
• Collected acceleration and angular velocity data during functional fitness drills.
• Trained and tested supervised learning models, including Support Vector Machine (SVM) with cubic kernel, using data windows.

Main Results:

• Achieved an overall classification accuracy of 97.8% using SVM with a cubic kernel and 600 ms data windows with 10% overlap.
• The best single-sensor configuration (upper arm) yielded 96.4% accuracy; two sensors (upper arm and thigh) achieved 97.6% accuracy.
• The method accurately distinguished between different drills but occasionally confused movement phases with transitions between repetitions.

Conclusions:

• Supervised learning effectively classifies complex sequential movements in functional fitness workouts.
• The developed approach, leveraging consumer-grade sensor technology, has significant potential for on-field applications.
• This technology can enable real-time feedback and analysis for unstructured training environments.