Sleep EEG Signal Analysis using Graph Embedding Simplicial Convolutional Recurrent Attention Network with Duck Swarm Algorithm

• Published in: 2025 3rd International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS) pp. 613 - 618
• Main Authors: Khrais, Ibrahim Mohammad, S, Sheela, Sethi, Gaurav, Salomi Victoria, D. Rosy, Sashmi, S.Nooray, Vijay, G.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 11.06.2025
• Subjects: Accuracy; AutoEncoder Filter Bank Common Spatial Patterns; Brain modeling; Classification algorithms; Convolution; Convolutional codes; Duck Swarm Algorithm; Electroencephalography; Graph Embedding Simplicial Convolutional Recurrent Attention Network; Heuristic algorithms; Python; Short-Time Fourier Transform and Continuous Wavelet Transform; Signal analysis; Sleep; Sleep EEG Analysis
• DOI: 10.1109/ICSSAS66150.2025.11080817

Sleep EEG Analysis (S-EEG-A) records brain activity to categorize sleep stages, identify patterns, and study disorders by frequency and waveform analysis. Sleep EEG signals are hard to analyze due to their complexity, noise, and high dimensionality. Temporal and spatial EEG patterns cannot be processed easily by conventional models, and they decrease classification accuracy and generalization. To that end, the present paper puts forward a Graph Embedding Simplicial Convolutional Recurrent Attention Network with a Duck Swarm Algorithm (GESCR2A2Nets+DSA) to analyze Sleep EEG Signal. Input data are accessed from the Sleep-EDF dataset and then preprocessed in the first step via the AutoEncoder Filter Bank Common Spatial Patterns (AE-FBCSP) process. Second-level feature extraction is then passed through the Short-Time Fourier Transform and Continuous Wavelet Transform (S-tFTCWT). Classification based on Graph Embedding Simplicial Convolutional Recurrent Attention Network (GESCR2A2Nets) and later optimized through Duck Swarm Algorithm (DSA). The performance of the resulting GESCR2A2Nets+DSA model is tested on a Sleep-EDF dataset at an accuracy of 99.9% and F1-score of 99.8%. The resulting model is constructed utilizing Python programming language. The performance of the suggested GESCR2A2Nets+DSA model demonstrates excellent accuracy and robust performance in sleep stage classification, well identifying complex EEG patterns and improving generalization across the Sleep-EDF dataset.