EEG Signal Classification Based on Multitask Diffusion Adaptation over Customized Hyper-Networks

The classification of multi-channel signals remains a challenging yet essential task in biomedical signal processing. Among these, electroencephalogram (EEG) signals stand out due to their potential to assist clinicians in diagnosing neurological conditions. Existing classification methods, however, often struggle with issues such as noise, variability across subjects, and the complexity of real-time data processing. This study introduces a novel classification framework grounded in the concept of a customized hyper-network. Within this architecture, hyper-nodes dynamically represent independent adaptive networks that collectively enhance the system's flexibility and robustness. By integrating a classification strategy into the hyper-network structure, the proposed method demonstrates improved performance in handling real EEG signals, even when corrupted by noise. Extensive comparative experiments conducted on both synthetic and real-world EEG datasets validate the proposed approach. The results highlight its superior accuracy and adaptability in comparison to existing methods, underscoring the framework's potential for real-time applications and clinical integration.