Advanced Two-Stage Authentication with EEG: Inter-Subject P300 ERP Classification Using EEGNet for User Recognition

A reliable biometric system is crucial in modern security challenges. EEG signals offer a promising alternative to conventional biometrics due to their uniqueness, forgery resistance, and long-term stability. Among EEG-based approaches, P300 event-related potentials (ERPs) are widely studied for their robustness in distinguishing individuals. This study proposes a two-stage authentication framework integrating user identification (multi-class classification) and authentication (binary classification) for enhanced security. A comprehensive EEG preprocessing pipeline including bandpass filtering, epoch rejection, temporal cropping, and downsampling was implemented to improve signal quality and extract cleaner P300 responses. EEGNet, a lightweight convolutional neural network designed for EEG analysis, was used for feature extraction and classification. 16 participants across three sessions evaluated the system robustness under cross-session conditions. The framework demonstrated strong generalization, achieving up to 95% precision and a mean accuracy of 96.10% for both identification and authentication. These results highlight the potential of EEGNet user authentication using P300 ERPs, offering insights into optimal preprocessing and deep learning methods for EEG-based security. This study underscores brainwave-driven authentication as a secure and viable alternative to traditional biometrics, paving the way for real-time applications in cybersecurity and identity verification.