An unsupervised subject identification technique using EEG signals.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalography (EEG) is a non-invasive method to record brain activity.
• Identifying individuals from EEG signals is challenging due to inherent variability.
• Unsupervised methods are desirable for subject identification without prior labeling.

Purpose of the Study:

• To develop an unsupervised method for subject identification using EEG spectral features.
• To map high-dimensional EEG spectral data into a low-dimensional space for visualization and analysis.
• To evaluate the efficiency of the proposed method in discriminating between different subjects.

Main Methods:

• Extraction of power spectral density (PSD) from EEG signals across different frequency bands.
• Application of t-distributed stochastic neighbor embedding (t-SNE) for non-linear dimensionality reduction to a 2D feature space.
• Utilizing fuzzy c-means clustering for unsupervised subject discrimination.

Main Results:

• EEG spectral features were successfully mapped into a visually distinct 2D feature space, forming well-separated clusters for each subject.
• The fuzzy c-means clustering algorithm effectively identified different subjects within the 2D space.
• The proposed method demonstrated high efficiency, achieving a precision greater than 90% in discriminating 10 subjects.

Conclusions:

• Unsupervised subject identification from EEG spectral features is feasible using non-linear dimensionality reduction and clustering.
• The proposed t-SNE and fuzzy c-means approach provides a robust and efficient method for EEG-based subject discrimination.
• This technique holds potential for applications requiring individual identification from brain activity patterns.