EEG Authentication System Based on One- and Multi-Class Machine Learning Classifiers.

Area of Science:

• Biometrics and Cybersecurity
• Machine Learning Applications
• Signal Processing

Background:

• Secure access to personal and professional information is crucial in the digital age.
• Traditional authentication methods face challenges in ensuring robust user verification.
• Biometric authentication offers personalized security by utilizing unique user characteristics.

Purpose of the Study:

• To propose and evaluate an electrocardiogram (ECG)-based user authentication system.
• To introduce Isolation Forest and Local Outlier Factor classifiers for biometric authentication using ECG data.
• To identify key ECG channels and brainwaves for authentication and compare dimensionality reduction techniques.

Main Methods:

• Implementation of One-Class and Multi-Class Machine Learning classifiers, specifically Isolation Forest and Local Outlier Factor.
• Analysis of Electrocardiogram (ECG) data for feature extraction and classification.
• Comparison with Principal Component Analysis (PCA) and χ2 statistical test for dimensionality reduction.

Main Results:

• Isolation Forest and Local Outlier Factor demonstrated suitability for ECG-based authentication.
• Identification of significant EEG channels and brainwaves contributing to authentication accuracy.
• A hybrid system combining Isolation Forest and Random Forest achieved 82.3% accuracy, 91.1% precision, and 75.3% recall.

Conclusions:

• ECG-based biometric authentication using novel machine learning classifiers is a viable security solution.
• The proposed hybrid system offers robust resistance against random forgery attacks.
• This research contributes to advancing personalized and secure user authentication methods.