Automated Identification of Epileptic Seizures from Electroencephalography Data using Deep Features and Metaheuristic-Driven Model

Visually recognizing the epileptiform segments in intricate electroencephalography (EEG) data is a difficult, laborious, and error-prone process. This has led to a surge in the popularity of machine learning (ML) and deep learning algorithms (DL) for quick and automatic seizure recognition. Existing works using ML need handcrafted features and sophisticated choice of feature ranking and classification methods. As a result, the output is highly influenced by the choice of attribute extraction, ranking, and classification methods. Conversely, DL approaches offer a complete end-toend framework by automatically identifying the characteristics required for the classification. Still, they are time-intensive and need more generalization. To address this issue, we propose an expert model for seizure detection based on integrating deep features and a BAT-enhanced Fuzzy K-nearest neighbor (FKNN) model. The paper improves upon our previous work, using the learned Alexnet model features for SVM model training. To further enhance the detection accuracy and motivated by the usage of the BAT optimization algorithm over the diverse classification applications, we propose its use along with FKNN for classifying the seizure events. The FKNN hyperparameters are optimized using BAT. The Bonn and the Neurology & Sleep Centre (NSC) datasets have been taken to validate the efficacy of the designed approach. The suggested approach attained a maximum level of performance across both datasets (sensitivity $=100 \%; \mathbf{9 8 \%}$, accuracy $=100 \%; \mathbf{9 8. 6 7 \%}$, & specificity $=100 \%; 100 \%$, which is higher than the existing works. The proposed automated tool has the implications of assisting the clinician in automatically marking the relevant segments for review, thereby saving time and improving diagnostic efficiency.