Sleep stages detection from EEG signal utilizing Backpropagation Neural Network and Deep Neural Network

Prompt and accurate identification of sleep-related disorders is essential for mitigating the progression to more serious neurodegenerative conditions. However, current diagnostic solutions, such as biomarkers and cognitive screening tests, are expensive, time-consuming, or not user-friendly. This research evaluated a Neural Network (NN) against a Deep Neural Network (DNN) in categorizing five sleep stages based on EEG data. To ensure high-quality input data for the model, we employ artifact correction, signal decomposition, and overlapping sliding window processing, then extracting time-domain, frequency-domain, and non-linear domain features. The evaluation process centered on assessing their performance using precision, recall, and F1-score metrics. Overall, the DNN outperformed the NN, especially in distinguishing the wake (W) and rapid eye movement (R) stages, highlighting its ability to capture subtle EEG patterns. Although the NN succeeded, particularly in classifying certain stages, it struggled with more complex distinctions, such as between stages N1, N2, and R. This comparison emphasizes the advantage of deeper network architectures like DNN for analyzing intricate EEG signals. However, the higher computational complexity of DNNs also presents challenges, suggesting the need for optimization in future studies to balance performance with efficiency.