Overcoming behavioral variability in electromyography signals by an adaptive incremental classification approach.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Surface electromyography (sEMG)-based gesture recognition systems face performance issues with physiological variations.
• Lifestyle factors like smoking and alcohol consumption cause distributional shifts, degrading model accuracy.

Purpose of the Study:

• To introduce an adaptive incremental k-Nearest Neighbors (ADINC-kNN) algorithm for robust sEMG gesture recognition.
• To enable dynamic adaptation to population-specific physiological variations without complete model retraining.

Main Methods:

• Implemented an adaptive incremental k-Nearest Neighbors (ADINC-kNN) algorithm.
• Utilized a sliding-window buffer and distance-weighted voting for dynamic decision boundary refinement.
• Evaluated on 15 hand force exercises with 14 subjects using 5-fold cross-validation.

Main Results:

• ADINC-kNN significantly outperformed static kNN in accuracy, precision, recall, and F1-score.
• Achieved over 90% classification performance in smoking and alcohol-consuming groups.
• Demonstrated a superior balance between computational efficiency and predictive accuracy compared to retraining-based methods.

Conclusions:

• ADINC-kNN offers a scalable and practical solution for robust sEMG gesture recognition.
• The algorithm effectively adapts to diverse user populations and changing physiological conditions.
• Suitable for real-world applications in rehabilitation, assistive technology, and human-machine interaction.