The field of brainwave-based biometrics has gained attention for its potential to revolutionize user authentication through hands-free interaction, resistance to shoulder surfing, continuous authentication, and revocability. However, current research often relies on single-session or limited-session datasets with fewer than 55 subjects, raising concerns about generalizability and robustness. To address this gap, we conducted a large-scale study using a public brainwave dataset of 345 subjects and over 6,000 sessions (averaging 17 per subject) recorded over five years with three headsets. Our results reveal that deep learning approaches outperform classic feature extraction methods by 16.4\% in Equal Error Rates (EER) and comparing features using a simple cosine distance metric outperforms binary classifiers, which require extra negative samples for training. We also observe EER degrades over time (e.g., 7.7\% after 1 day to 19.69\% after a year). Therefore, it is necessary to reinforce the enrollment set after successful login attempts. Moreover, we demonstrate that fewer brainwave measurement sensors can be used, with an acceptable increase in EER, which is necessary for transitioning from medical-grade to affordable consumer-grade devices. Finally, we compared our findings with prior work on brainwave authentication and industrial biometric standards. While our performance is comparable or superior to prior work through the use of Supervised Contrastive Learning, standards remain unmet. However, we project that achieving industrial standards will be possible by training the feature extractor with at least 1,500 subjects. Moreover, we open-sourced our analysis code to promote further research.
View on arXiv