Classification of cyclic alternating patterns of sleep using EEG signals

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Cyclic Alternating Patterns (CAP) are crucial EEG signal components during non-rapid eye movement sleep.
• Analyzing CAP provides insights into various sleep disorders.
• Accurate identification of CAP phases A and B is essential for detailed sleep analysis.

Purpose of the Study:

• To develop an easy-to-implement and accurate system for differentiating CAP phases A and B.
• To compare the performance of different machine learning classifiers for CAP classification.
• To establish a reliable method for real-time CAP analysis.

Main Methods:

• EEG signal segments processed using Gaussian filters to obtain sub-band components.
• Statistical features extracted from signal components.
• Minimum Redundancy Maximum Relevance (mRMR) test for feature selection.
• Comparison of three machine learning classifiers: k-nearest neighbor (kNN), and others.

Main Results:

• The k-nearest neighbor (kNN) classifier achieved 79.14% accuracy and 79.24% F-1 score on a balanced dataset.
• The proposed method demonstrated superior performance compared to existing CAP classification techniques.
• Analysis considered both balanced and unbalanced datasets for robust evaluation.

Conclusions:

• The developed system offers an accurate and straightforward method for CAP phase classification.
• The kNN classifier shows strong performance, making it suitable for sleep disorder analysis.
• The system's ease of implementation and accuracy position it as a candidate for real-time deployment in clinical settings.