A New Approach for Sleep Stage Identification Combining Hidden Markov Models and EEG Signal Processing

Abstract Purpose Sleep constitutes a third of human life, underscoring its importance in health-related and psychophysiological research. Monitoring sleep stage evolution is critical for understanding sleep-related issues and diagnosing disorders. This study aims to classify sleep stages using a Hidden Markov Model (HMM) based on spectral statistical measures derived from raw electroencephalography (EEG) signals. It explores effective feature combinations to enhance classification accuracy while maintaining a practical approach requiring minimal inputs. Methods We utilized raw EEG signals to extract various statistical features in the frequency domain, identifying combinations that maximize predictive performance. The proposed HMM was employed to classify sleep stages, leveraging these spectral features. Unlike many prior studies that focus solely on machine learning (ML) techniques, our analysis emphasizes feature significance and model interpretability. Results Our approach achieved a multiclass classification accuracy of 76.76% using only EEG recordings. This performance demonstrates the utility of spectral statistical features for sleep stage classification, with results comparable to more complex ML methods. Conclusion The proposed methodology highlights a practical, accurate and interpretable approach to sleep stage classification using EEG data. Its simplicity and efficiency make it suitable for both offline and online applications, supporting improved diagnosis of sleep disorders and advancing sleep research.