An effective hybrid feature selection using entropy weight method for automatic sleep staging.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Sleep Medicine

Background:

• Manual sleep staging is subjective and time-consuming.
• Automatic sleep staging using machine learning shows promise but is hindered by high-dimensional, redundant features.
• Effective feature selection is crucial for improving automatic sleep staging accuracy.

Purpose of the Study:

• To propose a hybrid feature selection method for automatic sleep staging.
• To address the issue of redundant and irrelevant features in polysomnography (PSG) data.
• To enhance the accuracy and efficiency of sleep stage classification.

Main Methods:

• Preprocessing of four-modal PSG signals (EEG, EOG, ECG, EMG).
• Extraction of 185 time, frequency, and nonlinear features.
• A two-stage hybrid feature selection: Entropy Weight Method combined with filter methods, followed by Sequential Forward Selection.
• Ensemble classification model using SVM, KNN, Random Forest, and MLP.

Main Results:

• The hybrid method selected 30 highly relevant features for sleep staging.
• Achieved 88.86% accuracy, 83.15% F1 score, and 0.8531 Kappa coefficient for 6-class sleep staging.
• Demonstrated superior performance compared to existing state-of-the-art methods.

Conclusions:

• The proposed hybrid feature selection method is effective for automatic sleep staging.
• Significant reduction in feature dimensionality while maintaining high classification performance.
• Offers a promising approach for objective sleep quality assessment and diagnosis of sleep disorders.