Experimental studies on the efficiency of people authentication using EEG data from the same and different examinations

The paper presents the results of preliminary studies on the authentication of human subjects using electroencephalography (EEG) signals. Many research results in the literature assume that the data used both for training and testing may come from the same set. In this work, we verified how the authentication results change if the data used for training and testing are from separated recording sessions. The examinations used in the studies were collected from 36 healthy adults and 20 EEG sessions were planned for each person. We evaluated the statistical metrics of the considered authentication methods based on the analysis of signal features (parameters) from different EEG bands: separately for delta, theta, alpha, beta, and gamma, as well as for the combinations of features from the different EEG bands. In the first approach, the first 15 examinations from each participant were used for training and the remaining 5 for testing. In the second approach, 75 % of features from each session were used for training and the remaining 25 % for testing. The results show that the statistical metrics of authentication analysis obtained using signals from the same set, both for training and testing, turned out to be better than when we split the data using individual sets for training and examination. In practice, the authentication is performed based on a current examination, different from the ones used for training. Therefore, for the proper evaluation of the authentication methods, only the metrics obtained when the examinations used for training and testing are separated should be used. The results also suggest that there are significant statistical differences in accuracy and sensitivity between the two approaches.