Motor imagery EEG classification with optimal subset of wavelet based common spatial pattern and kernel extreme...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor imagery brain-computer interfaces (MI BCIs) performance relies heavily on feature extraction.
• Filter-bank common spatial pattern (FBCSP) methods increase features but risk overfitting in MI BCIs.
• Existing methods face challenges in balancing feature richness with model robustness.

Purpose of the Study:

• To improve MI BCI performance by addressing feature extraction challenges.
• To mitigate overfitting issues associated with high-dimensional feature sets in MI BCIs.
• To enhance the computational efficiency of MI BCIs.

Main Methods:

• Implemented an eigenvector centrality feature selection method.
• Utilized wavelet packet decomposition common spatial pattern (WPD-CSP).
• Employed a kernel extreme learning machine (KELM) classifier.

Main Results:

• The proposed methods successfully improved MI BCI performance.
• Overfitting problems were effectively avoided with the new feature selection and classification approach.
• Kernel extreme learning machine significantly enhanced computational speed.

Conclusions:

• The combination of eigenvector centrality, WPD-CSP, and KELM offers a robust solution for MI BCIs.
• This approach enhances both accuracy and efficiency, overcoming limitations of traditional FBCSP methods.
• The findings pave the way for more practical and high-performing brain-computer interfaces.