Employing Biomedical Signals as a Means of Biometric Identification

Abstract

In modern society, different identification and verification methods are being used, and everyone finds that security is a top concern. Traditional methods like passwords and hardware tokens may be lost or stolen, resulting in identification failure. Therefore, we require reliable and robust human recognition techniques. Using electroencephalography (EEG), person identification systems can be robust, anti- spoof, and efficient. This paper aims to develop an efficient deep-learning model for person identification using EEG signals. We proposed a deep learning model using a 1D Convolutional Neural Network (CNN) and stacked Long Short-Term Memory (LSTM) separately. The effective dataset DEAP was used for evaluating the proposed method. The proposed models were trained on the first 10 seconds of EEG data (60 seconds long) and tested on random 10 seconds of data from the remaining part of the data. The results indicate that stacked LSTM slightly outperformed 1D CNN with up to 99.97% accuracy with just eight data channels (out of 64 channels) and 15 subjects. The comparative analysis between parameters like the number of channels used, the length of data used for training and testing, and the number of subjects indicates that stacked LSTM outperforms 1D CNN as RNNs can remember certain features of time-series data better than CNN.