A Comprehensive Review of Machine Learning Techniques in Sleep Staging Systems

Sleep staging is a critical process in diagnosing and understanding various sleep disorders. Traditional manual scoring methods are time-consuming, subjective, and require significant expertise, prompting the exploration of automated systems powered by machine learning (ML). This paper presents a comprehensive analysis of ML techniques employed in sleep staging systems, focusing on their architectures, features, datasets, and performance metrics. We review various models, including traditional ML approaches. Key studies highlight innovations like feature extraction from multi-modal signals, including EEG, EMG, EOG, and respiratory data, to enhance classification accuracy across different sleep stages. Challenges such as dataset quality, generalization, interpretability, and computational cost are discussed, alongside recent advancements addressing these issues. The analysis underscores the potential of ML in automating sleep staging with high accuracy and efficiency while emphasizing the need for standardized datasets, interpretable models, and robust validation frameworks.