Multi-paradigm EEG signals and AI enable reliable biometric authentication

Electroencephalography, the recording of the brain’s electrical activity through electrodes on the scalp, has long been viewed as a promising biometric modality. EEG is considered a promising biometric modality because of the inherent uniqueness of neural signals and their resistance to spoofing attacks. A new study identifies a key obstacle to reliable EEG-based authentication: brain signals are highly non-stationary and vary across cognitive states, so a signature captured in one cognitive condition may not hold in another.

To address this, researchers developed a multi-paradigm measurement protocol designed to capture stable, subject-specific EEG signatures across diverse brain conditions. Rather than recording from a single task, the protocol gathers signals during six complementary paradigms: eyes-open and eyes-closed resting states, left- and right-hand motor tasks, working-memory encoding, steady-state visual evoked potentials, and steady-state auditory evoked potentials. The aim is to build an identity template that holds up regardless of what the brain is doing at the moment of authentication.

The recorded signals were preprocessed and segmented into two-second epochs with 50 percent overlap. From each channel the team extracted 28 quantitative descriptors spanning four domains: time, frequency, time-frequency, and nonlinear features. This feature set was then fed to several decision models, including Mahalanobis distance-based template matching, logistic regression, support vector machines, random forests, and a one-dimensional convolutional neural network.

Performance was assessed under both single-paradigm and multi-paradigm configurations using cross-session validation, a demanding test because it requires the system to recognize individuals across separate recording sessions. The multi-paradigm protocol reduced the equal error rate by 2 percent compared with single-paradigm systems, indicating that combining diverse cognitive states yields more stable neural measurements than any one state alone.

The best results came from a surprisingly simple model. Logistic regression achieved an equal error rate of 1.6 percent, with a 95 percent confidence interval of 1.48 to 1.73. At a false acceptance rate of 1 percent, the false rejection rate was 2.06 percent; tightening security to a 0.1 percent false acceptance rate raised the false rejection rate to 9.01 percent. These figures suggest a workable trade-off between convenience and security, though the confidence interval and the cross-session design should be kept in view when interpreting the headline number.

Feature analysis pointed to which signals carry identity information most reliably.

The study’s authors conclude that multi-paradigm EEG acquisition improves the stability of neural measurements across cognitive states and enables reliable biometric authentication in dynamic conditions.

Several caveats remain. The reported results come from a controlled measurement protocol evaluated with cross-session validation, and the paper is being shared early as a peer-reviewed, accepted version subject to further edits. Nonetheless, the finding that mixing cognitive paradigms lowers error rates offers a concrete design principle for EEG biometrics: identity lives not in a single brain state but in how a person’s brain behaves across many.

Subject of Research: EEG-based biometric authentication using multi-paradigm neural recordings and machine learning decision models

Article Title: EEG-biometric authentication system using multi-paradigm neurophysiological measurements and artificial intelligence

Article References: Rakshe, C., Kumar, P. S., Thomas, C. B., Kalathe, M., Sreeraj, V. S., Venkatasubramanian, G., Thomas, J., Rajamanickam, Y., Kumar, D., & Ronickom, J. F. A. (2026). EEG-biometric authentication system using multi-paradigm neurophysiological measurements and artificial intelligence. Brain Informatics. https://doi.org/10.1186/s40708-026-00337-7

Image Credits: AI Generated

DOI: 10.1186/s40708-026-00337-7

Keywords: EEG, biometric authentication, electroencephalography, machine learning, logistic regression, equal error rate, neural signals, cross-session validation, Hjorth mobility, SSVEP, SSAEP, convolutional neural network