Detection of Epilepsy Seizure in Adults Using Discrete Wavelet Transform and Cluster Nearest Neighborhood Classifier

• Published in: Iranian journal of science and technology. Transactions of electrical engineering Vol. 45; no. 4; pp. 1103 - 1115
• Main Authors: Syed Rafiammal, S., Najumnissa Jamal, D., Kaja Mohideen, S.
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01-12-2021; Springer Nature B.V
• Subjects: Accuracy; Adults; Algorithms; Classification; Classifiers; Clusters; Complexity; Convulsions & seizures; Decomposition; Discrete Wavelet Transform; EEG; Electrical Engineering; Electroencephalography; Engineering; Epilepsy; Feature extraction; Feature selection; Frequencies; Learning algorithms; Machine learning; Neighborhoods; Research Paper; Seizures; Temporal variations; Wavelet transforms; Feature selection; Seizure; KNN; Discrete wavelet transform; Machine learning
• ISSN: 2228-6179; 2364-1827
• DOI: 10.1007/s40998-021-00437-6

Seizure detection from EEG signal plays important role in diagnosing and treating the Epilepsy disease. Development of Low complexity detection algorithms is needed in order to design efficient automatic epilepsy detection devices. In this paper, an automatic seizure detection algorithm proposed using Discrete Wavelet Transform and Cluster-based Nearest Neighborhood machine learning algorithm. The Electroencephalogram signals decomposed by Daubechies Wavelet transform. Temporal features extracted from decomposed Wavelet sub-bands. A new distance-based feature selection method introduced for an optimal feature selection. The proposed Cluster-based KNN algorithm reduces the number of computations required for conventional KNN method. The performance of proposed algorithm is validated by publically available benchmark EEG database. This proposed Classification method obtained 100% accuracy between seizure and normal EEG signals; 98% of accuracy between Inter-ictal and seizure signals, 91% of accuracy between Normal and Inter-ictal signals. This proposed cluster nearest neighborhood classifier requires less number of training samples and less number of calculation steps to detect seizure events. The analysis on classification performance between the various frequency bands confirms that, the EEG signal frequency band of 2.6–5.5 Hz reveals better classification results in adults. Due to less complexity of algorithm, the proposed algorithm is well suited for hardware implementation of automatic seizure detection systems.