EEG feature selection and the use of Lyapunov exponents for EEG-based biometrics

In recent years, there has been a growing increase in the use of electroencephalographic (EEG) signals for biometric systems. In investigating the use of EEG-based biometrics in a smart-device environment, this study focused on the development of a specific feature selection method, and on the feasibility of nonlinear dynamic characteristics of EEG signals for identifying individuals. We recorded sixteen EEG channel signals from seven subjects during two minutes in resting state with eyes closed, for a minimum of five times over several days. Power spectral density and the maximum Lyapunov exponents were calculated for the individual EEG characteristics. A specific criteria index (CI) that consisted of three types of variances was developed to quantify the level of EEG features, and to select adequate feature candidates with not only a low intra-subject variability but also high inter-subject discrimination. Statistical t-tests and a preliminary classification test using a linear support vector machine (SVM) classifier quantified the performance of feature selection, giving an accuracy rate of 94.9% for identifying each individual. In addition, they also revealed that the maximum Lyapunov exponents are one of the most feasible features for an EEG-biometric system, with an accuracy rate of 85.5% when using only maximum Lyapunov exponents from two EEG channels (T4, F4).