Using Hidden Markov Model for Identification Based on EEG Signals

Abstract

The researches on individual identification approaches based on EEG signals draw lots of attention in recent years. Few of them got time-robust identification performance. In this study, we focused on the time robustness of individual identification using EEG under conditions of resting-state of eye open/closed (REO/REC). Ten subjects participated in this study and each of them conducted three independent runs experiment, with the time intervals between adjacent runs were at least two weeks. There were three sessions within each run, and the time duration of each session is 150 seconds of REO/REC. Two features, auto-regressive (AR) and Mel-frequency cepstrum coefficients (MFCC) were calculated as identity features. Then Support vector machines (SVM) and Hidden Markov model (HMM) were used as classifiers. To access the time-robust performance of our methods, we used one of three runs data as test set and the other two as training set. Results show that the best classification accuracy is 80%. It is believed that under the conditions of REC and REO, the identity features of most subjects are robust across time and can be used for identification. This study will have an important impact on in EEG-based identification system.