Exploring Feature Selection and Classification Techniques to Improve the Performance of an...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Classifying motor imagery (MI) activities using electroencephalogram (EEG) signals is crucial for brain-computer interfaces (BCIs).
• Existing BCI systems struggle with accuracy due to non-discriminative and ineffective features extracted from EEG data.
• Developing robust feature extraction and selection methods is essential for improving BCI performance.

Purpose of the Study:

• To propose a novel multiband decomposition-based feature extraction and selection method for enhanced MI classification in BCIs.
• To address the challenge of high dimensionality and ineffectiveness of features in current EEG-based BCI systems.
• To improve the accuracy and reliability of motor imagery recognition for BCI applications.

Main Methods:

• Preprocessing EEG signals and decomposing them into four sub-bands.
• Applying Common Spatial Pattern (CSP) technique for narrowband feature extraction within each sub-band.
• Utilizing the Relief-F algorithm for effective feature selection to reduce dimensionality, followed by advanced classification.

Main Results:

• The proposed method demonstrated superior performance in classifying MI tasks across three benchmark EEG datasets.
• Achieved higher classification accuracy compared to existing state-of-the-art BCI systems.
• Successfully reduced feature dimensionality while enhancing the discriminative power of selected features.

Conclusions:

• The developed multiband decomposition and feature selection approach significantly improves MI classification accuracy in BCIs.
• The method offers an effective solution for handling complex EEG data and overcoming limitations of traditional feature extraction techniques.
• This work contributes to the advancement of more reliable and efficient brain-computer interfaces for individuals with motor impairments.