Automated Sleep Staging System with EEG Signal using Machine Learning Techniques

Advances in signal processing and machine learning have revolutionized the analysis of physiological signals, providing much better results as compared to traditional manual methodology. This paper focuses on automatic sleep stage classification via EEG channels Fpz and Cz extracted from the publicly available SleepEDF PSG Hypnogram dataset. Those channels were preprocessed to improve their quality and mapped the bio-signals to corresponding sleep stages, which is a decisive step for accurate classification to be achieved. It utilized the sleep stage scoring through various machine learning approaches, including Random Forest, Gradient Boosting, Bagging Classifier, and an Ensemble Learning approach. This yielded classification accuracies as high as 78% with Random Forest, 79% with Gradient Boosting, 75% with Bagging Classifier, and 85% with Ensemble Learning last one being the most promising model. This work highlights the relevance of EEG channels in classifying sleep stages and demonstrates how machine learning can be exploited to move the analysis of sleep data forward. Implications from these findings go towards embedding automated sleep stage classification in consumer-grade sleep monitoring systems that would contribute towards more efficient and accessible sleep health monitoring solutions.