Improving classification accuracy of motor imagery EEG using genetic feature selection

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor imagery (MI) classification is crucial for brain-computer interfaces (BCI).
• Accurate feature extraction and selection are vital for improving MI classification performance.
• Existing methods may not fully optimize feature sets for robust MI decoding.

Purpose of the Study:

• To propose an electroencephalogram (EEG) analysis system incorporating feature selection to enhance motor imagery (MI) data classification.
• To evaluate the effectiveness of a genetic algorithm for selecting optimal features from extracted EEG data.
• To compare the performance of the proposed system against traditional methods and assess its suitability for BCI applications.

Main Methods:

• Extraction of diverse EEG features: adaptive autoregressive (AAR) parameters, spectral power, asymmetry ratio, coherence, and phase-locking value.
• Application of a genetic algorithm for selecting the most relevant features from the extracted set.
• Classification of selected features using a support vector machine (SVM) classifier.

Main Results:

• The proposed system demonstrated superior classification accuracy for MI data compared to analyses without feature selection.
• The SVM classifier, utilizing genetically selected features, outperformed the back-propagation neural network (BPNN) on two independent datasets.
• The feature selection process significantly contributed to the enhanced classification performance.

Conclusions:

• The developed EEG analysis system with feature selection effectively improves motor imagery classification accuracy.
• The integration of genetic algorithms for feature selection and SVM for classification is a promising approach for BCI development.
• The proposed system shows significant potential for real-world brain-computer interface applications requiring reliable MI decoding.