Sliding-Window Normalization to Improve the Performance of Machine-Learning Models for Real-Time Motion Prediction...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Electromyography (EMG) signals are crucial for controlling advanced prosthetic devices and assistive technologies.
• High classification accuracy is essential for reliable EMG-based motion prediction.
• Traditional normalization methods like z-score require calibration, limiting their real-time application in EMG studies.

Purpose of the Study:

• To develop and validate a novel, calibration-free normalization method for EMG signals.
• To improve the real-time classification accuracy of machine learning models for EMG-based motion prediction.
• To assess the effectiveness of the proposed method in improving cross-subject and real-time EMG signal processing.

Main Methods:

• Proposed a new normalization technique combining sliding-window and z-score normalization for real-time EMG processing.
• Implemented and tested the method on single-joint elbow movement (rest, flexion, extension) prediction.
• Evaluated performance with and without calibration, including cross-subject data application.

Main Results:

• The proposed normalization method achieved 77.7% accuracy, a significant improvement over non-normalized data (56.2%).
• In a cross-subject application without calibration, the method reached 63.1% accuracy, outperforming standard z-score normalization (54.4%).
• Demonstrated the method's effectiveness in enhancing classification performance for EMG-based motion prediction.

Conclusions:

• The developed sliding-window and z-score normalization is a simple, effective, and calibration-free solution for real-time EMG signal processing.
• This method substantially improves the accuracy of machine learning models for EMG-based motion prediction.
• The findings pave the way for more robust and user-friendly EMG-controlled assistive devices.