ABSTRACT
This work proposes a novel approach for identification of epilepsy from electroencephalogram (EEG) data obtained from both healthy subjects as well as affecred patients. EEG data contains crucial information reflecting various physiological states of the brain. Our method involves employing hierarchical clustering techniques for seizure detection. We have evaluated several models for this task, including CNN+SVM, VGG16, DenseNet121, MobileNetV2, and Xception. Remarkably, DenseNet121 achieved 100% accuracy, followed closely by VGG16 with 99.98%, and MobileNetV2 with 99.95%, while Xception attained 98.07% accuracy and the CNN+SVM model achieved 94.98%. Based on these results, we selected DenseNet121 as the optimal model for building our frontend application using Flask. The utilization of DenseNet121 in the front end provides a robust foundation for real-time seizure detection, offering potential benefits for both medical professionals and patients. Employing pre-trained models accelerates the convergence process during fine-tuning, leading to a reduction in the time and computational resources needed for training.
