The application of mutual information-based feature selection and fuzzy LS-SVM-based classifier in motion...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electromyography (EMG) signals are crucial for motion classification.
• Existing feature reduction methods can be computationally intensive and may compromise accuracy.

Purpose of the Study:

• To develop an effective mutual information-based feature selection approach for EMG-based motion classification.
• To reduce computational complexity while maintaining high classification accuracy.

Main Methods:

• Wavelet packet transform (WPT) was used to decompose EMG signals into sub-bands.
• Mutual Information (MI) theory was applied for feature reduction.
• Compared MI-based selection with PCA, SFS, and BE.
• Evaluated performance using fuzzy least squares support vector machines (LS-SVMs) and neural networks (NN).

Main Results:

• The MI-based feature selection demonstrated superiority over PCA, SFS, and BE in terms of speed and accuracy.
• The combination of MI-based feature selection and SVM techniques outperformed PCA and NN.
• High accuracy was achieved in identifying diverse motions using MI and SVM.

Conclusions:

• The proposed filter-based feature extraction and reduction strategy is effective and classifier-independent.
• SVM techniques combined with WPT and MI show significant potential for EMG motion classification.