BMT_EEG: A Novel EEG Dataset for Evaluating the Effects of New Protocols on Biometric Authentication Systems

Due to uniqueness, revocability, and resilience to theft, Electroencephalography (EEG) biometric solutions are becoming more popular for securing high stake systems. To propel research in EEG-based biometrics, we introduced a novel dataset: Biometric EEG Dataset (BMT_EEG), uniquely designed to address anti-spoof concerns of Biometric applications. The dataset captures neural responses of 23 users across three sessions, encompassing eleven protocols that include seven motor movement (MM) and motor imagery (MI) tasks, two visually evoked potentials (VEPs), and two resting-state conditions. Leveraging BMT_EEG, we employed EEGNet, a compact convolutional neural network designed for EEG analysis, and a baseline CNN-GRU model combining 1-D Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) for modeling spatio-temporal dependencies for user authentication. Both the models were evaluated under random and skilled forgery attacks in inter-session setting, using two sessions for training and the third for testing, enhancing robustness against session variability and its suitability for real-world deployment scenario. The CNN-GRU model achieved the best average Equal Error Rate (EER) of 1.29 ± 0.92% for random forgery using the MI2 task and 6.15 ± 3.78% for skilled forgery using the MI1 task. Notably, EEGNet, achieved better performance with an EER of 0.06±0.04% for random forgery using MI2, and 2.14±4.58% for skilled forgery using MI1. This work underscores the importance of realistic evaluation protocols, comprehensive EEG dataset, and suitability of deep learning architectures in advancing biometric research.