Hybrid Convolutional Autoencoder Models for EEG-Based Epileptic Seizure Detection

Epileptic seizure is a widespread disease among many other neurological disorders, and its diagnosis requires precise and efficient detection. Our study employs a novel hybrid CAE framework that integrates feature extraction with optimized classification methods, reducing dependence on manual preprocessing. The proposed CAE architecture is particularly suitable for the Bonn EEG dataset, which contains five specific classes, achieving an accuracy of 98.7% for seizure detection. It enabled the system to obtain a sensitivity of 98.9% and a specificity of 98.5%, supporting consistent detection with minimal false alarms. The overall training time is reduced by 35% compared to traditional systems, and the adaptive pooling mechanism has enhanced the model’s generalizability. Equipped with these capabilities, the hybrid CAE framework is a reliable solution for real-time seizure detection and can be integrated into portable monitoring devices.