Ant colony optimization-based feature selection method for surface electromyography signals classification.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Surface electromyography (sEMG) signals are crucial for classifying human motion.
• High-dimensional sEMG features pose challenges for accurate and efficient classification.
• Feature selection is essential for optimizing classification performance.

Purpose of the Study:

• To propose a novel Ant Colony Optimization (ACO) based feature selection method for sEMG signals.
• To enhance the accuracy and computational efficiency of hand motion classification.
• To validate the proposed method against Principal Component Analysis (PCA).

Main Methods:

• Developed an ACO-based feature selection scheme incorporating the Minimum Redundancy Maximum Relevance (mRMR) criterion (ACO-mRMR).
• Extracted two feature sets: Time Domain features with Autoregressive model coefficients (TDAR) and Wavelet Transform (WT) features.
• Compared ACO-mRMR with PCA for feature reduction on sEMG data from ten subjects performing eight upper limb motions.

Main Results:

• ACO-mRMR achieved average classification accuracies of 95.45±2.2% for TDAR features and 96.08±3.3% for WT features.
• PCA resulted in lower average accuracies: 91.51±4.9% for TDAR and 89.87±4.4% for WT.
• The proposed ACO-mRMR method demonstrated superior performance in sEMG classification.

Conclusions:

• The ACO-mRMR feature selection method significantly improves classification accuracy for sEMG-based motion classification.
• This approach is computationally efficient and effective for handling high-dimensional sEMG data.
• The method shows potential applicability for pattern analysis in other biomedical signals.