Fine-Tuning EEG Channel Utilization for Emotionally Stimulated Biometric Authentication

Biometric authentication relies on distinct biological traits of individuals to validate their identity, enhancing security measures. However, variations in an individual’s emotional state can impact the reliability of the biometric system. In this study, we propose a novel pipeline to evaluate electroencephalography (EEG)-based biometric system across different emotional states and optimize critical brain regions using machine learning algorithms. EEG signals from the DEAP dataset were classified into four emotional states: HAHV, HALV, LALV, and LAHV. We extracted a comprehensive set of statistical, time, frequency, entropy, fractal, spectral, and shape features from each channel. Machine learning classifiers, including Random Forest, Gradient Boosting, Extreme Gradient Boosting, LightGBM, CatBoost, and Bagging, were used for participant authentication. Our results revealed that the CatBoost classifier performed well across all stimuli with average accuracies of 84%, 85%, 86%, and 83% for HAHV, HALV, LALV, and LAHV, respectively. We found that features from channels FC1, Fz, C4 & Pz, and FC1 significantly contributed to EEG authentication on stimuli such as HAHV, HALV, LALV, and LAHV, respectively. Features such as skewness and the theta-to-alpha frequency band ratio consistently performed well across stimuli, demonstrating EEG signals’ potential for robust biometric authentication by addressing emotional variations.