Multi-Feature Fusion Approach for Epileptic Seizure Detection From EEG Signals

Abstract

In this article, a new fusion scheme based on the Dempster-Shafer Evidence Theory (DSET) is introduced for Epileptic Seizure Detection (ESD) in brain disorders. Firstly, various features in temporal, spectral, and temporal-spectral domains are extracted from Electroencephalogram (EEG) signals. Afterward, a Correlation analysis via the Pearson Correlation Coefficient (PCC) is conducted on the extracted features to select and remove highly correlated features. It leads to the second feature set with about half numbers of the first feature set. Next, three separate filter-type feature selection techniques, including Relief-F (RF), Compensation Distance Evaluation Technique (CDET), and Fisher Score (FS), are conducted to this second feature set for ranking features. Following that, a feature fusion is engaged by the DSET through the individual feature ranking results to generate high qualified feature sets. Indeed, the DSET-based feature fusion is devoted to enhancing the feature selection confidence using the least superb ranked features. In the classification stage, an Ensemble Decision Tree (EDT) classifier, along with two common validation procedures, including hold out and 10-fold cross-validation, is appropriated to classify the selected features from the EEG signals as normal, pre-ictal (epileptic background), and ictal (epileptic seizure) classes. Finally, several test scenarios are investigated using experimental data of Bonn University to evaluate the proposed ESD performance. Moreover, a comparison with other research works on the same dataset and classes is accomplished. The obtained results indicate the effectiveness of the proposed feature fusion approach and superior accuracy compared to the traditional methods.