Methods of Selecting Electroencephalographic Features for Personal Authentication

Abstract

Biometric technology based on electroencephalography (EEG) identifies individuals by using personal characteristics in human brainwaves. This study aims to evaluate EEG features and channels for biometrics and to propose a methodology that selects the optimal features to discriminate individuals. Thirty healthy subjects participated in the study. While recording EEG signals from fourteen channels, subjects were asked to relax and keep eyes closed for two minutes. To evaluate intra-individual variability, we recorded EEG ten times for each subject across different days to reduce any within-day effect. From each channel, eight EEG features were calculated including alpha/beta power ratio, alpha/theta power ratio, beta/theta power ratio, median frequency, PSD entropy, permutation entropy, sample entropy, and maximum Lyapunov exponents. These features were evaluated by three feature selection algorithms based on Fisher score, reliefF, and information gain, respectively. A linear discriminant analysis (LDA) classifier along with a leave-one-out cross validation method discriminated self against others using the selected features. The best feature set was found to be composed of 23 features highly ranked on Fisher scores, which yielded a 18.56% half total error rate. In addition, EEG channels located on occipital and right temporal areas appeared to most contribute to authenticate individuals.