Strategic Feature Extraction for Improved Seizure Detection: A Tanh and LeakyReLU Activated Neural Network Model

With epilepsy affecting millions worldwide, a robust seizure detection model is paramount for advancing patient care. The methodology utilizes a dataset comprising five classes, each representing different brain activities, and employs a meticulous preprocessing strategy for classification between seizure and non-seizure instances. The proposed model’s architecture is a key highlight, featuring parallel dense blocks and concatenation layers for hierarchical and diverse feature extraction. The strategic use of Tanh and LeakyReLU activations enhances the model’s adaptability to the complex nature of EEG signals. Results showcase the model’s exceptional accuracy, achieving 98.97% on the training data and 97.57% on the testing data. This research contributes to a high-performing seizure detection model and provides insights into the significance of parallel architectures and strategic feature extraction in enhancing model capabilities. The results position the proposed model as a promising tool for integration into clinical decision support systems.