Leveraging Multiple Distinct EEG Training Sessions for Improvement of Spectral-Based Biometric Verification Results

Area of Science:

• Neuroscience
• Biometrics
• Signal Processing

Background:

• Electroencephalography (EEG)-based biometrics often rely on limited data, potentially overestimating performance due to signal variability.
• EEG signals are susceptible to interferences, electrode placement, and transient states, impacting recognition accuracy.

Purpose of the Study:

• To investigate the impact of varying numbers of distinct EEG recording sessions on the performance of EEG-based verification.
• To determine the optimal number of training sessions for robust EEG biometric systems.

Main Methods:

• Analyzed EEG data from 29 participants (20 sessions each) and 23 impostors (1 session each).
• Utilized raw and decibel-scaled power spectral density coefficients as input for a shallow neural network.
• Evaluated system performance based on accuracy, sensitivity, and impostor attack rates across different training/testing session counts.

Main Results:

• The number of distinct recording sessions significantly affects the sensitivity of EEG-based verification.
• Performance gains plateaued when using more than eight training sessions under the study's conditions.
• Achieved 96.7 ± 4.2% accuracy with 15 training sessions; 15 training sessions resulted in a 3.1 ± 2.2% impostor attack rate.

Conclusions:

• Multiple recording sessions are crucial for training reliable EEG-based biometric systems.
• Increasing the number of test sessions did not significantly enhance results.
• Findings suggest a baseline for EEG recognition, emphasizing multi-session data for training robustness.