Abstract
Sleep staging is essential for evaluating sleep quality, diagnosing disorders, and creating personalized treatment plans. The convolutional and bidirectional long short-term memory hybrid neural network (CNN-BiLSTM) has shown promise in automated sleep staging from electroencephalogram (EEG) signals. However, prior studies often overlook expert-derived manual features, relying solely on deep neural networks for automatic feature extraction. This study proposes an automated sleep staging model (named MA-CNN-BiLSTM) for single-channel EEG using a CNN-BiLSTM network with embedded manual features and attention mechanisms. The model computes multidimensional features such as signal energy and entropy via wavelet decomposition and integrates attention mechanisms to enable the network to focus on crucial features for classification. Sleep stage classification is achieved using a SoftMax layer. The proposed MA-CNN-BiLSTM model is validated on the Sleep-EDF-20 and SVUH-UCD datasets, demonstrating superior classification accuracy, macro-averaged F1 scores, and Cohen's Kappa, outperforming other models.