Automated Sleep Stage Classification Using Biorthogonal Wavelet Decomposition and Machine Learning Techniques
Authors :- P Verma, H Telangore, M Sharma, D Joshi
Publication :- International Conference on Artificial Intelligence and Machine Vision (AIMV), IEEE, 2025.
Diagnosing sleep disorders like insomnia, sleep apnea, and narcolepsy involves assessing sleep. Adequate sleep is also necessary for overall health upkeep. Since manual ways of sleep stage categorization are time-consuming and susceptible to human error, automatic means are preferred. In this research, an automated system for sleep stage classification via machine learning algorithms and signal processing methods is illustrated utilizing the Wisconsin Sleep Cohort (WSC) dataset, which was obtained from the National Sleep Research Resource (NSRR). Our physiological signals from polysomnography (PSG) recordings consist of electroencephalograms (EEG) with channels C3-M2 and O1-M2, electrooculograms (EOG) with channels E1 and E2, and electromyograms (EMG) with a single channel lleg-r. Sleep stages and 30-second epochs are assigned to the signals according to the provided hypnogram. Feature extraction uses five-level time-frequency biorthogonal wavelet decomposition to produce six sub-bands per signal. Hjorth parameters—activity, mobility, and complexity—are calculated for all sub-bands, resulting in 90 features. Classified features are then used by many machine learning models, of which the top performer is the ensemble bagged tree (EBT) classifier. It registers a performance of 86.2% accuracy at 200 Hz. All of these findings show that the suggested technique presents a secure and scalable mechanism for the sleep stage classification task, towards developing automated sleep analysis.