Machine learning-based EEG signals classification model for epileptic seizure detection

• Published in: Multimedia tools and applications Vol. 80; no. 12; pp. 17849 - 17877
• Main Authors: Aayesha, Qureshi, Muhammad Bilal, Afzaal, Muhammad, Qureshi, Muhammad Shuaib, Fayaz, Muhammad
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2021; Springer Nature B.V
• Subjects: Algorithms; Classification; Computer Communication Networks; Computer Science; Convulsions & seizures; Data Structures and Information Theory; Datasets; EEG; Electroencephalography; Epilepsy; Feature extraction; Machine learning; Multimedia Information Systems; Seizure detection; Seizures; Signal classification; Signal processing; Special Purpose and Application-Based Systems; Signal processing; Seizure detection; Epilepsy; EEG; Machine learning; Classification
• ISSN: 1380-7501; 1573-7721; 1573-7721
• DOI: 10.1007/s11042-021-10597-6

The detection of epileptic seizures by classifying electroencephalography (EEG) signals into ictal and interictal classes is a demanding challenge, because it identifies the seizure and seizure-free states of an epileptic patient. In previous works, several machine learning-based strategies were introduced to investigate and interpret EEG signals for the purpose of their accurate classification. However, non-linear and non-stationary characteristics of EEG signals make it complicated to get complete information about these dynamic biomedical signals. In order to address this issue, this paper focuses on extracting the most discriminating and distinguishing features of seizure EEG recordings to develop an approach that employs both fuzzy-based and traditional machine learning algorithms for epileptic seizure detection. The proposed framework classifies unknown EEG signal segments into ictal and interictal classes. The model is validated using empirical evaluation on two benchmark datasets, namely the Bonn and Children’s Hospital of Boston-Massachusetts Institute of Technology (CHB-MIT) datasets. The obtained results show that in both cases, K-Nearest Neighbor (KNN) and Fuzzy Rough Nearest Neighbor (FRNN) give the highest classification accuracy scores, with improved sensitivity and specificity percentages.