A Novel Methodology for Classifying EMG Movements Based on SVM and Genetic Algorithms

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electromyography (EMG) signal processing is crucial for medical research, enabling non-invasive diagnosis, treatment, and rehabilitation.
• EMG signals present challenges due to their random, non-stationary, and non-linear nature, complicating classification.
• Identifying key features is vital for improving EMG signal classification accuracy.

Purpose of the Study:

• To propose a genetic algorithm-based methodology for feature selection in EMG signal processing.
• To identify the optimal feature subset that minimizes classification error in EMG signal segments.
• To compare the effectiveness of genetic algorithms with particle swarm optimization (PSO) for EMG feature selection.

Main Methods:

• Utilized genetic algorithms for feature selection to optimize the parameter space for EMG signal classification.
• Employed a support vector machine (SVM) for the classification task.
• Applied the methodology to two databases: right upper extremity and right lower extremity movements.
• Implemented Particle Swarm Optimization (PSO) for comparative analysis on the upper extremity dataset.

Main Results:

• Achieved over 65% feature space reduction for both upper and lower extremity datasets.
• Attained an average classification efficiency of 91% for the best feature subset identified by genetic algorithms.
• PSO resulted in an 88% average error and 46% feature reduction for the upper extremity data.
• Sensitivity analysis indicated that features selected by genetic algorithms demonstrated greater sensitivity in the classification process.

Conclusions:

• Genetic algorithms provide an effective approach for feature selection in EMG signal processing, significantly improving classification accuracy.
• The proposed methodology offers a robust method for enhancing the development of EMG-based diagnostic and rehabilitation devices.
• Genetic algorithms outperform PSO in terms of feature selection sensitivity and classification performance for EMG data.