A robust principal component analysis algorithm for EEG-based vigilance estimation.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalogram (EEG) data is high-dimensional and noisy, requiring effective feature reduction.
• Robustness in feature dimensionality reduction is crucial for accurate EEG data utilization.

Purpose of the Study:

• To introduce and evaluate a robust Principal Component Analysis (PCA) algorithm for EEG feature dimension reduction.
• To compare the performance of robust PCA against standard PCA, L1-norm PCA, and sparse PCA for vigilance estimation.

Main Methods:

• Utilized smoothed differential entropy features as vigilance-related EEG features.
• Applied robust PCA and compared its performance with standard PCA, L1-norm PCA, and sparse PCA on a 23-subject EEG dataset.
• Evaluated algorithms for both off-line and on-line vigilance estimation.

Main Results:

• Robust PCA demonstrated superior robustness and performance compared to other tested algorithms.
• Vigilance estimation using robust PCA achieved an average Root Mean Square Error (RMSE) of 0.158.
• Standard PCA resulted in a higher average RMSE of 0.172 for vigilance estimation.

Conclusions:

• Robust PCA is an effective method for reducing the dimensionality of EEG features for vigilance estimation.
• The proposed robust PCA algorithm offers improved accuracy and reliability in vigilance monitoring compared to traditional PCA methods.