abstract
- Neural based subject authentication has emerged as a promising paradigm in biometric security, with the P300 event related potential (ERP) gaining particular attention due to its reliability and individual specificity. The proposed scheme introduces a personalized brainprint recognition framework that utilizes an enhanced AMEEGNet architecture to accurately classify symbol evoked P300 responses. A visual oddball paradigm featuring directional arrows and a stop sign was employed to elicit P300 activity with minimal cognitive demand. 16 subjects (participants) completed the task across three separate sessions conducted on different days. EEG data from the session 1 and 2 were used to train the deep learning model, while the third session was utilized for cross-session validation and testing. The proposed framework achieved strong performance, with an average classification accuracy of 89.25% and average Cohen¿s kappa score of 0.88. These findings demonstrate that symbol-evoked P300 responses carry consistent, individual-specific features, underscoring their potential for use in practical EEG-based subject authentication systems.