Machine Learning based Epileptic Seizure Detection from EEG Signal: Bangladesh Perspective

Epileptic seizures (ES) pose significant challenges to individuals' well-being and quality of life, with a global impact affecting millions of people. Detecting and classifying ES is essential for proper medical intervention and management. This study addresses ES detection using EEG analysis, focusing specifically on the context of Bangladesh. This study presents a EEG-based method for detecting epileptic seizures differentiating between seizure and non-seizure signals. Utilizing empirical mode decomposition, essential subcomponents are extracted from EEG signals, from which three distinct features were derived. Neural Network (NN), the well-known machine learning model, is employed for ES classification from the extracted features. The significance of this research lies in the utilization of real-life EEG data from a hospital in Khulna, Bangladesh, rather than relying solely on publicly available benchmark datasets. Processing real-life EEG data presents unique challenges, including data preprocessing and variability in EEG signals due to different devices and channel configurations. The proposed model achieved remarkable average accuracies, with 91.05% using 50 hidden neurons in the NN. To validate the effectiveness of the proposed method, outcomes were also compared with publicly available benchmark dataset. The proposed method offers a reliable and efficient approach for ES detection, particularly relevant for real-life healthcare settings in Bangladesh, contributing to improved epilepsy management.