Feature selection using mutual information for EEG-based biometrics

Abstract

Recently, electroencephalography (EEG) has emerged as a novel means to identify an individual for biometric authentication. Successful application of EEG to biometrics relies on how well the signal features of EEG represent individual identities. In this study, we propose a new approach to the selection of an optimal EEG feature set, using a mutual information technique. The EEG data were recorded with 21 dry electrodes from 7 subjects while they rested with eyes closed for 2 minutes. Seven features (alpha/theta, alpha/beta, theta/beta power ratio, sample entropy, permutation entropy, entropy, and median values of distribution) were calculated for each EEG channel, and mutual information between each pair of features was calculated for each subject. Then we selected the optimal features that exhibited the largest intra-subject mutual information. Using the selected features, we performed an authentication test by means of a Bhattacharyya distance-based nearest-neighbor method with leave-one-out cross-validation. As a result, with best nine features we achieved a 95% accuracy rate. Our results suggest a feasibility of using a mutual-information-based feature selection scheme for EEG-based biometrics.