Deep Learning Approaches for EEG-Based Biometrics: A Systematic Review

Biometics such as fingerprint, face, and iris are vulnerable to spoof attacks. The unique characteristics of Electroencephalography (EEG) make it a promising biometric modality especially because of its resistance to spoofing attacks. Many deep learning methods have been proposed for EEG-based biometric systems. This systematic review examines these methods in terms of their feature extraction ability and authentication performance. We follow the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines to search IEEE Xplore, PubMed,Web of Science, ScienceDirect, and Springer databases. Initially, we identified 285 relevant articles published between 2018 and 2024. After removing duplicates and applying predefined inclusion and exclusion criteria, 34 of them were selected and investigated thoroughly. This review highlights that deep learning architectures such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Generative Adversarial Networks (GANs), and Transformers have been applied in EEG biometric systems. These architectures are capable of automatically extracting discriminative features from raw EEG signals without the need for feature engineering. They achieved a high identification or verification performance with over 90% accuracy and a low Equal Error Rate (EER). However, challenges persist with these models, such as generalizability and adaptability across diverse datasets and subjects, and the long-term stability of EEG signatures over time. These challenges underscore the need for more robust and generalizable solutions. This review consolidates EEG-based deep learning developments, identifies common trends and gaps, and provides insights into the current capabilities and limitations of EEG biometrics, thereby serving as a comprehensive reference for this field. Key future directions are proposed, including the design of hybrid deep learning architectures, leveraging transfer learning techniques, and conducting rigorous cross-subject evaluations to further improve the generalizability and real-world reliability of EEG-based biometric systems.