Classification of EEG Using Adaptive SVM Classifier with CSP and Online Recursive Independent Component Analysis

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Traditional Common Spatial Patterns (CSP) for electroencephalography (EEG) feature extraction are susceptible to artifacts and non-stationary signals.
• Existing CSP methods lack frequency domain information and necessitate numerous input channels, limiting their application in real-time brain-computer interfaces (BCIs).
• Support Vector Machines (SVMs) generally offer superior classification performance compared to other methods in BCI applications.

Purpose of the Study:

• To develop an advanced feature extraction technique for EEG signals that overcomes the limitations of traditional CSP.
• To integrate frequency domain information and improve robustness against artifacts and non-stationarity in EEG data.
• To enhance the classification accuracy of online and real-time BCIs using a novel feature extraction and classification approach.

Main Methods:

• Proposed a novel feature extraction method combining Online Recursive Independent Component Analysis (ORICA) with Common Spatial Patterns (CSP), termed ORICA-CSP.
• Employed an Adaptive Support Vector Machine (A-SVM) classifier for enhanced performance in EEG data classification.
• Evaluated the ORICA-CSP method using a 4-class motor imagery dataset (Dataset 2a of BCI Competition IV).

Main Results:

• The proposed ORICA-CSP method demonstrated promising results in extracting features from EEG data.
• The integration of ORICA-CSP with A-SVM showed potential for improving classification accuracy in BCI applications.
• The method addresses limitations of traditional CSP, including artifact sensitivity and lack of frequency domain analysis.

Conclusions:

• The ORICA-CSP feature extraction method offers a significant improvement over traditional CSP for EEG-based BCIs.
• The combined ORICA-CSP and A-SVM approach shows effectiveness for online and real-time BCI applications.
• This study provides a robust and efficient method for enhancing BCI performance, particularly in challenging real-world scenarios.