Feature selection for EEG Based biometrics

Abstract

EEG-based biometrics identify individuals by using personal and distinctive information in human brain. This thesis aims to evaluate the electroencephalography (EEG) features and channels for biometrics and to propose methodology that identifies individuals. In my research, I recorded fourteen EEG channel signals from thirty subjects. While record EEG signal, subjects were asked to relax and keep eyes closed for 2 minutes. In addition, to evaluate intra-individual variability, we recorded EEG ten times for each subject, and every recording conducted on different days to reduce within-day effects. After acquisition of data, for each channel, I calculated eight features: alpha/beta power ratio, alpha/theta power ratio, beta/theta power ratio, median frequency, PSD entropy, permutation entropy, sample entropy, and maximum Lyapunov exponents. Then, I scored 112 features with three feature selection algorithms: Fisher score, reliefF, and information gain. Finally, I classified EEG data using a linear discriminant analysis (LDA) with a leave-one-out cross validation method. As a result, the best feature set was composed of 23 features that highly ranked on Fisher score and yielded a 18.56% half total error rate. In addition, according to scores calculated by feature selection, EEG channels located on occipital and right temporal areas most contributed to identify individuals. Thus, with suggested methodologies and channels, implementation of efficient EEG-based biometrics is possible.