Deep Learning-Based Sleep Stage Classification: An Eeg Signal Analysis Method

Abstract

Sleep stage identification is an important task in clinical assessment and neuroscience research. Manual scoring of electroencephalogram (EEG) recordings is time-consuming, resource-intensive, and subject to inter-scorer variability. This study presents a deep learning-based framework for automated sleep stage classification using EEG signals from the Sleep-EDF Expanded dataset. The proposed pipeline combines signal preprocessing, artifact correction, Continuous Wavelet Transform (CWT)-based time-frequency representation, and a lightweight convolutional neural network (CNN). EEG recordings are segmented into 30-second epochs and classified into the five standard sleep stages: Wake, N1, N2, N3, and REM, following AASM scoring rules. Preprocessing includes Butterworth band-pass filtering, notch filtering, z-score normalization, and artifact detection with interpolation-based correction. CWT scalograms are resized to 128×128 and supplied to a CNN containing two convolutional blocks followed by dense layers and a Softmax output. The model is trained using Adam optimization and cross-entropy loss with subject-level separation and 5-fold cross-validation. The reported experimental results show an overall accuracy of 88.4%, F1-score of 0.86, and Cohen's Kappa of 0.82. The framework provides a compact end-to-end approach for EEG-based sleep staging and is designed with computational efficiency and potential real-time or wearable deployment in mind.