Improving the separability of motor imagery EEG signals using a cross correlation-based least square support vector...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCIs) show rapid development but face challenges in motor imagery (MI) signal classification.
• Accurate classification of electroencephalogram (EEG) signals is crucial for effective BCI performance.

Purpose of the Study:

• To propose and validate a hybrid algorithm for improving the classification accuracy of MI-based EEG signals in BCIs.
• To introduce a novel cross-correlation based feature extractor combined with a least square support vector machine (LS-SVM).

Main Methods:

• A hybrid algorithm combining a novel cross-correlation feature extractor with a least square support vector machine (LS-SVM) classifier.
• Comparison with logistic regression and kernel logistic regression classifiers using the same extracted features.
• Evaluation on BCI Competition III datasets (IVa and IVb) using 10-fold cross-validation.

Main Results:

• The proposed LS-SVM classifier demonstrated superior performance over logistic regression and kernel logistic regression.
• The hybrid approach achieved a 7.40% improvement in classification accuracy compared to the best of eight recently reported algorithms.
• Experimental results confirmed the effectiveness of the proposed method on two distinct BCI datasets.

Conclusions:

• The proposed hybrid algorithm significantly enhances the classification accuracy of MI-EEG signals for BCIs.
• The novel cross-correlation feature extraction combined with LS-SVM offers a promising direction for BCI development.
• This approach represents a substantial advancement over existing methods in MI signal classification.