Sleep Stage Classification

Sleep stage classification labels short segments of overnight recordings, usually polysomnography, as wake, N1, N2, N3, or REM to measure sleep architecture and support screening and research. Common approaches learn signal features with convolutional networks or spectral filters, then model transitions across epochs with recurrent networks, Transformers, state-space models, or sequence decoders; other work combines EEG with EOG, ECG, PPG, or motion and learns reusable representations from large PSG collections. Benchmarks such as Sleep-EDF, SHHS, and ISRUC test accuracy and agreement, while newer evaluations emphasize difficult N1 transitions, cross-cohort transfer, noisy labels, wearable signals, and energy use. Strong benchmark performance does not guarantee reliable clinical deployment: populations and recording channels vary, expert labels disagree, and finer-grained or multimodal predictions must be balanced against interpretability and efficient operation.