UMAP for Dimensionality Reduction in Sleep Stage Classification Using EEG Data.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Sleep is crucial for overall health and well-being.
• Electroencephalography (EEG) is a key technology for sleep research.
• Accurate sleep stage classification (N1-N3, REM) is essential for sleep analysis.

Purpose of the Study:

• To investigate the effectiveness of Uniform Manifold Approximation and Projection (UMAP) for EEG feature extraction in sleep studies.
• To evaluate UMAP's dimensionality reduction capabilities on EEG signals for improved sleep detection and analysis.
• To compare UMAP's performance against traditional band power analysis for sleep stage classification.

Main Methods:

• Applied Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction on EEG signal features.
• Utilized sleep stage classification as the primary analytical task.
• Compared classification accuracy and reliability metrics between UMAP and traditional band power analysis.

Main Results:

• UMAP demonstrated superior accuracy and reliability for sleep stage classification compared to traditional methods.
• An average 11% increase in accuracy and a 20% increase in Macro-F1 Score were observed using UMAP.
• The wakefulness stage showed a notable 23% increase in Macro-F1 Score with UMAP, highlighting its effectiveness.

Conclusions:

• UMAP is a powerful tool for reducing dimensionality in EEG data while preserving crucial information for sleep analysis.
• The 2D visualization capabilities of UMAP effectively cluster EEG signals, aiding in the understanding of sleep dynamics.
• UMAP offers a significant advancement in EEG-based sleep detection and classification, outperforming conventional techniques.