An explainable EEG epilepsy detection model using friend pattern

Abstract

The electroencephalography (EEG) signals are very important for obtaining information from the brain, and EEG signals are one of the cheapest methods to gather information from the brain. EEG signals have commonly been used to detect epilepsy. Therefore, the main objective of this research is to demonstrate the epilepsy detection capability of the presented new-generation relation-centric feature extraction function. In this research, we have presented a new-generation EEG signal classification model, and this model is an explainable feature engineering (XFE) model. To present this XFE model, a feature extraction function, termed friend pattern (FriendPat), has been introduced. The presented FriendPat is a distance- and voting-based feature extraction function. By deploying the introduced FriendPat, features have been extracted. The generated features have been selected using a cumulative and iterative feature selector, and the selected features have been classified using a t algorithm-based k-nearest neighbors (tkNN) classifier. By using channel information and Directed Lobish's (DLob) look-up table based on the brain cap used, DLob symbols have been generated, and these symbols create the DLob string for artifact classification. By using this generated DLob string and statistical analysis, explainable results have been obtained. To investigate the classification performance of the presented FriendPat XFE model, we have used a publicly available EEG epilepsy detection dataset. The presented model attained 99.61% and 79.92% classification accuracies using 10-fold cross-validation (CV) and leave-one-subject-out (LOSO) CV, respectively. This XFE model generates a connectome diagram for epilepsy detection.