Combination of heterogeneous EEG feature extraction methods and stacked sequential learning for sleep stage...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Accurate sleep stage classification is crucial for diagnosing sleep disorders.
• Traditional methods rely on manual scoring or simpler algorithms, which can be time-consuming and subjective.
• Electroencephalogram (EEG) signals contain rich information about brain activity during sleep.

Purpose of the Study:

• To develop an improved methodology for automated sleep stage classification.
• To enhance classification accuracy by combining diverse feature extraction techniques and advanced learning models.
• To leverage temporal dependencies between sleep stages for more robust predictions.

Main Methods:

• Feature extraction from electroencephalogram (EEG) signals using Hjorth features, wavelet transformation, and symbolic representation.
• Application of feature selection to identify the most relevant extracted features.
• Implementation of stacked sequential learning, utilizing a second-layer classifier with predicted sleep stages as input.

Main Results:

• Both combined feature extraction and stacked sequential learning significantly improved sleep stage classification accuracy.
• The proposed methodology achieved results closer to the consensus of expert sleep stage scoring.
• Feature selection effectively identified discriminative features for improved model performance.

Conclusions:

• The integration of multiple EEG feature extraction methods enhances the information captured for classification.
• Stacked sequential learning effectively incorporates contextual information from adjacent sleep stages, boosting accuracy.
• The proposed methodology offers a promising approach for accurate and automated sleep stage classification.