Robust Single-Trial EEG-Based Authentication Achieved with a 2-Stage Classifier.

Area of Science:

• Biometrics and Cybersecurity
• Neuroscience and Machine Learning

Background:

• Increasing risks of personal data exposure necessitate advanced security measures.
• Biometric authentication utilizes unique physiological and behavioral traits for identification.
• Electroencephalography (EEG)-based authentication leverages subject-specific brain responses for access control.

Purpose of the Study:

• To propose and evaluate a novel two-stage EEG-based authentication system.
• To compare the efficacy of deep learning (Inception, VGG) and Principal Component Analysis (PCA) for feature extraction.
• To assess the performance of a Support Vector Machine (SVM) classifier in authenticating subjects.

Main Methods:

• EEG signals were recorded from 105 subjects during a motor paradigm.
• Feature extraction was performed using Inception-like and VGG-like neural networks, and PCA.
• A Support Vector Machine (SVM) was employed for binary classification in the second stage.

Main Results:

• The VGG-like NN-SVM decoder achieved up to 90.68% accuracy with a 2.89% False Acceptance Rate (FAR) using 64 channels.
• The Inception-like NN-SVM decoder reached 93.40% accuracy with a 1.27% FAR using 64 channels.
• The PCA-SVM decoder demonstrated superior performance, achieving 95.64% accuracy with a 1.26% FAR using 64 channels.

Conclusions:

• EEG-based authentication systems can provide effective personal data security.
• Both deep learning and PCA show promise for EEG feature extraction in authentication.
• The PCA-SVM approach demonstrated the highest accuracy and lowest FAR in this study.