Neutrosophic similarity score-based entropy measure for focal and nonfocal electroencephalogram signal classification

Abstract An electroencephalogram (EEG) is a useful tool that measures the change in electrical activity of the human brain. EEGs can be used for identification of various brain disorders such as epilepsy. Partial epilepsy, which affects some portions of the brain, is called the epileptogenic area. While the epileptogenic area is used for recording the focal (F)-EEG signals, the nonfocal (NF)-EEG signals are recorded from other regions of the brain. In this chapter, we propose an approach for accurate discrimination of F-EEG and NF-EEG signals. The proposed approach initially converts EEG signals into color images by using the time-frequency (TF) transformation. More specifically, the short time Fourier transform (STFT) is considered for TF representation of the input EEG signals. The obtained TF color images are transferred into the neutrosophic set (NS) domain. Neutrosophy is a branch of philosophy that deals with indeterminate and inconsistent information. The neutrosophy produces truth (T), false (F), and indeterminacy (I) membership triplets that are used to form the neutrosophic similarity score function. This function is then applied on each color channel of the time-frequency images. A sliding window is used on each color channel to calculate the local entropy. The calculated entropies are concatenated for obtaining the feature vector. Various classifiers such as support vector machines (SVM), decision trees (DT), ensemble learners (EL), and k-nearest neighbors (k-NN) are used for classification. The obtained results are evaluated based on accuracy and compared with some existing results. The 99.8% accuracy scores are obtained with both the k-NN and ensemble-based methods. The evaluations show that the proposed approach is encouraging for future works.