An Autonomous Sleep-Stage Detection Technique in Disruptive Technology Environment.

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Accurate sleep stage classification is vital for addressing sleep disorders.
• Home-based autonomous sleep tracking requires reliable methods for sleep analysis.
• Traditional machine learning models often face limitations in performance and efficiency.

Purpose of the Study:

• To introduce a novel hybrid model, PSO-XGBoost, for enhanced sleep stage classification.
• To leverage Particle Swarm Optimization (PSO) for hyperparameter tuning of the Extreme Gradient Boosting (XGBoost) model.
• To evaluate the model's performance using electroencephalogram (EEG) signals for potential real-time sleep monitoring applications.

Main Methods:

• Feature extraction from EEG signals across time, frequency, and time-frequency domains.
• Implementation of a hybrid PSO-XGBoost model for sleep stage classification.
• Validation using the Pz-oz signal dataset from the sleep-EDF expanded repository with stratified K-fold cross-validation.

Main Results:

• Achieved high performance metrics: 95.4% accuracy, 95.4% F1-score, 95.4% precision, and 94.3% recall.
• Demonstrated an average accuracy of 95%, outperforming traditional machine learning methods.
• Identified prefrontal EEG derivations as optimal and highlighted the effectiveness of a feature-shifting approach.

Conclusions:

• The PSO-XGBoost model offers a computationally efficient and accurate solution for sleep stage classification.
• The findings support the use of wearable EEG devices with dry electrodes for feasible home-based sleep monitoring.
• The proposed method shows significant potential for real-time sleep classification applications.