PERFORMANCE COMPARISON OF SCHIZOPHRENIA DETECTION USING EEG SIGNALS

Abstract

Schizophrenia is a kind of mental illness that affects about 1% of the world's population. It may be difficult for schizophrenic patients to distinguish between internally generated and externally generated stimuli due to one or more problems with the nervous system's corollary discharge process. This is one explanation for some symptoms of schizophrenia. As a result, looking into this process and how it relates to the disease's symptoms could help us learn more about the abnormal brain processes seen in patients with this disease. New avenues in the study of electrophysiological brain activity can be explored, and ambulatory neuronal disease diagnosis can be performed, thanks to improved access to EEG data. The aim is to use a variety of diagnostic techniques to identify neuronal pathologies. The dataset used to test the methods was a larger sample replication of EEG data from previous studies published on July 10, 2013 in Schizophrenia Bulletin Advance Access. The electroencephalogram (EEG) data of 22 healthy people and 36 people with schizophrenia were combined with the EEG data of 10 healthy people and 13 schizophrenia patients from a previous study. Two different methods for diagnosing SZ using EEG signals were considered during the classification phase. To classify EEG signals, traditional machine learning techniques have been used. Among the methods used were xgboost, decision tree, naïve bayes, random forest, Long short-term memories (LSTMs), support vector machines, two-dimensional convolutional networks (2D-CNNs), and two dimensional convolutional networks-LSTMs. The Deep Learning models were implemented at this point, and a variety of activation functions were compared. Among all proposed models, the SVM architecture has demonstrated the highest level of performance. The RBF Kernel with Cross Validation(CV) = 3, 6, 8 are used in this architecture. A precision of 100 percent is achieved using the SVM model.