Hybrid Deep Variational Empirical Mode Decomposition for Robust EEG Seizure Detection and Time-Frequency Analysis
Ghasem Farjamnia *
Institute of Applied Mathematics, Baku State University, Baku, Azerbaijan.
Mehrdad Hashemi
Department of Applied Mathematics and Cybernetics, Baku State University, Baku, Azerbaijan.
*Author to whom correspondence should be addressed.
Abstract
Accurate detection of epileptic seizures from EEG signals remains a central challenge in biomedical signal processing due to the non-linear and non-stationary nature of brain activity. This study introduces a Deep Variational Empirical Mode Decomposition (DVEMD) framework that integrates variational mode initialization, empirical sifting, and adaptive noise control to generate noise-resilient and physiologically interpretable modes. DVEMD is evaluated on the University of Bonn EEG dataset and compared with EMD, EEMD, CEEMDAN, EWT, and VMD. Experimental results demonstrate that DVEMD achieves superior performance, reaching 99.2% accuracy, 98.8% sensitivity, and 99.4% specificity, improving classification accuracy by approximately 2–6% over traditional decomposition methods. The approach also offers strong potential for real-time implementation due to its computationally efficient hybrid design. These findings highlight DVEMD as a robust and interpretable solution for automated seizure detection and clinical EEG analysis.
Keywords: EEG, epilepsy seizure detection, hybrid decomposition, time-frequency analysis, deep learning, adaptive signal processing, deep variational empirical mode decomposition