Multi Channel EEG Based Biometric System with a Custom Designed Convolutional Neural Network

Abstract

In this study, a convolutional neural network (CNN) is designed to identify multi-channel raw electroencephalograph (EEG) signals obtained from different subjects. The dataset contains 14 channel EEG signals taken from 21 subjects with their eyes closed at a resting state for 120 s with 12 different stimuli. The resting state EEG waves were selected due to better performance in classification. For the classification, a Convolutional Neural Network (CNN) was custom designed to offer the best performance. With the sliding window approach, the signals were separated into overlapping 5 s windows for training CNN better. fivefold cross-validation was used to increase the generalization ability of the network. It has been observed that, while the proposed CNN is found to give a correct classification rate (CCR) of 72.71%, the CCR reached the level of average 83.51% by using 4 channels. Also, this reduced the training time from 626 to 306 s. Therefore, the results show that usage of specific channels increases the classification accuracy and reduces the time required for training.