Robust Epileptic Seizure Recognition using Dimensionality Reduction with Deep Learning on EEG Signals

An epileptic seizure is a sudden surge of electrical activity in the brain, disrupting normal brain function and often resulting in loss of consciousness or convulsions. The most important diagnostic test for epilepsy is the Electroencephalogram (EEG). Usually, the recognition of epileptic activity is based on finding specific patterns in the multimodal EEG and is done by the human expert. This is a time-consuming and difficult process; therefore, numerous attempts have been made to automate it using both Deep Learning (DL) and conventional methods. Epileptic seizure detection using DL includes training neural networks for analyzing the EEG signal and detecting patterns indicative of seizures with a high level of accuracy for earlier
diagnosis and treatment. This study introduces a Robust Epileptic Seizure Recognition using Metaheuristics-based
Dimensionality Reduction with Deep Learning (RESR-MDRDL) technique on EEG signals. The RESR-MDRDL technique
concentrates on accurately identifying epileptic seizures utilizing EEG signals. In a preliminary stage, the RESR-MDRDL technique performs data pre-processing to standardize the input data. Also, a Salp Swarm Algorithm (SSA)-based technique is utilized for optimum Feature Selection (FS). For seizure recognition, the RESR-MDRDL technique employs a Deep Autoencoder (DAE) model, and its efficiency is improved using a Tunicate Swarm Algorithm (TSA). The simulation of the RESR-MDRDL methodology is examined by using an EEG dataset. The experimental validation of the RESR-MDRDL methodology indicated a superior accuracy value of 94.14% over existing techniques.