Title: Multiclass epilepsy seizure classification using deep learning

Authors: Pankaj Kunekar; Mukesh Kumar Gupta; Pramod Gaur

Addresses: Swami Keshvanand Institute of Technology, Management & Gramothan, Jaipur, 302017, India ' Swami Keshvanand Institute of Technology, Management & Gramothan, Jaipur, 302017, India ' BITS Pillani, Dubai Campus, UAE

Abstract: The neurosurgeon must be able to identify the type of epileptic seizure for the purpose to comprehend the cortical connectivity of the brain. Recurrent seizures, often known as epilepsy, are symptoms of the central nervous system and last either a few seconds or, in rare cases, a few minutes. EEG is one of the methods for recording seizures. The majority of EEG devices are composed of scalp electrodes that capture electrical activity. These signals are complex and often difficult to classify. Despite the existence of automated premature seizure identification from a usual electroencephalogram, little effort at multiple classes classification of seizures have been made done. Therefore, utilising the Bonn University dataset, a deep learning models has been developed using RNN, LSTM, and bi-directional-long short-term memory also known as Bi-LSTM, to solve this challenge. Bi-LSTM is found to be the most effective model for multi-class categorisation of epilepsy episodes in this research, with an accuracy of nearly 99% for three classes.

Keywords: epilepsy seizures; electroencephalogram; EEG; Bi-LSTM; deep learning.

DOI: 10.1504/IJMEI.2025.149555

International Journal of Medical Engineering and Informatics, 2025 Vol.17 No.6, pp.597 - 610

Received: 29 Jun 2023
Accepted: 25 Sep 2023
Published online: 07 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article