The Role of Multiple Data Characteristics in EEG-Based Biometric Recognition: The Impact of States, Channels, and Frequencies

Abstract

Electroencephalogram (EEG)-based biometric recognition systems have attracted substantial interest in recent years because of their promising applications in various fields, notably in biometric security. This study examines the influence of multiple attributes of data, such as channels, frequencies, and data states, on the accuracy of person recognition using EEG data. We analyzed EEG recordings from 20 participants, captured with affective and non-affective, that contain resting, visually evoked potential, and motor activity states, during a single session. Utilizing a data set comprising both affective and non-affective states, we developed a person recognition system that is resistant to affective fluctuations and the challenges posed by non-affective conditions. This hybrid data set (combining data on affective and non-affective states) introduces a novel approach that extends beyond what has been addressed in previous studies. Furthermore, our research establishes a comparative framework for evaluating the effects of alpha, beta, and gamma frequency ranges, as well as frontal, central, and occipital channel regions, on the biometric system. This framework offers valuable insights into which factors perform better for recognition throughout affective and non-affective data sets. The results obtained by employing a surrogate model demonstrated accuracy rates of 95.34% for affective, 84.86% for non-affective, and 93.92% for hybrid data sets. Notably, the affective data demonstrated higher classification success across all EEG frequencies and channel regions compared to the non-affective data.