Gender, Age and Number of Participants Effects on Identification Ability of EEG-based Shallow Classifiers

Biometric systems have numerous advantages over traditional authentication methods and should be taken into account: uniqueness, fraud-resistance, portability, convenience of use, scalability, possibility of reliable attendance recording. Use of electroencephalography (EEG) data as biometric factor can keep all these benefits and add new ones. This paper provides insight how EEG based biometrics identification ability can be affected by the number of participants, their gender and age. In order to study aforementioned factor's influence Leipzig Study for Mind-Body-Emotion Interactions (LEMON) resting state (both eyes-open state and eyes-closed state) EEG dataset were used. After few preprocessing adjustments and applying discrete wavelet transform as time-frequency domain feature extraction method resulting dataset were used for classification task. The testing was conducted in 5-fold cross-validation mode using the shallow model, Linear Support Vector Classifier. From experiment results following conclusions were made: increased number of participants decreased classifier accuracy, with age increase accuracy steadily improved, gender effects on accuracy are inconclusive.