Abstract

Epilepsy is considered one of the common medical and social disorders with unique characteristics. EEG signal was used for the classification and detection of epilepsy. This study proposed epilepsy classification without signal decomposition, as well as other algorithms used for decomposing the EEG signal to sub-bands like discrete wavelet transform (DWT) and dual-tree complex wavelet transform (DT-CWT). Descriptive comparisons were done between results for EEG signals with/without decomposition. The proposed algorithm includes the study of the extracted features and using machine learning kernels as Support Vector Machine (SVM) and bagged tree to achieve the optimal values of (accuracy-specificity-sensitivity and execution time). Results show that adding the line length to the group of features, the accuracy increased to 99.4%. By employing decomposing the EEG signal, the accuracy could be raised to99.875 % even after reducing the number of features to only three features. These features are line length, STD, and mean. This study proposed different algorithms with minimum features for epilepsy classification and localization with optimum execution time.