Automatic Sleep Stage Classification from Single-Lead ECG

This project explores the use of single-lead ECG signals for automated sleep stage classification, motivated by the need for accessible and wearable-compatible alternatives to traditional polysomnography (PSG). Sleep plays a vital role in health and recovery, and understanding its architecture - comprising Wake, REM, Light, and Deep stages - is essential for diagnosing and managing sleep disorders. However, conventional sleep monitoring requires EEG-based lab setups that are expensive, intrusive, and impractical for everyday use. I was particularly interested in heart rate variability (HRV) as a non-invasive proxy for autonomic nervous system activity during sleep. Prior research has shown that HRV patterns shift meaningfully across stages, making ECG a powerful yet underutilized signal for sleep analysis. This led me to design an end-to-end classification framework centered around physiologically interpretable HRV features, with the goal of enabling lightweight and continuous sleep monitoring from everyday biosensors like smartwatches and rings.