Multi-Classification of Electroencephalogram Epileptic Seizures Based on Robust Hybrid Feature Extraction Technique and Optimized Support Vector Machine Classifier

Epilepsy is a disease with various forms. However, limited dataset has confined classification studies of epilepsy into binary classes only. This study sort to achieve multiclassification of epileptic seizures through a robust feature extraction technique by comprehensively analyzing various advanced feature parameters from different domains, such as energy and entropy. The values of these parameters were computed from the coeficients of dilation wavelet transform (DWT) and modified DWT, known as dual-tree complex wavelet transform decomposition. The model was evaluated from the features of each of the parameters. The hybrid features were divided into three experiments to extract the meaningful features as follows: 1). features from combined energy features were extracted; 2). features from combined entropy features were also extracted; and 3). features from combined parameters as hybrid features were extracted. Finally, the model was developed based on the extracted features to perform a multi-classification of seven types of seizures using an optimized support vector machine (SVM) classifer. A recently released temple university hospital corpus dataset consisting of long-time seizure recordings of various seizures was employed to evaluate our proposed model. The proposed optimized SVM classifer with the hybrid features performed better than other experimented models with the value of accuracy, sensitivity, specificity, precision, and F1-score of 96.9%, 96.8%, 93.4%, 95.6%, and 96.2%, respectively. The developed model was also compared with some recent works in literature that employed the same dataset and found that our model outperformed all the compared studies. Index Terms--Multi-classification, EEG, epileptic seizures, DTCWT, hybrid features, SVM