Channel selection for EEG-based biometric recognition

Abstract

Person identification is an important factor for information systems. Emerging technologies for security such as biometric identification based on EEG signals, although promising, still require extensive research and further refinement before applied in practice. This work address the problem of EEG channel selection for biometric identification. Five forms of EEG signal segmentation are explored before the feature extraction by autoregressive model (AR model). Channel selection is performed with two approaches, genetic algorithms and search-based ranking, and we use the classifiers k-nearest neighbor (KNN) and support vector machines (SVM) for identification. The results indicate that it is possible to decrease up to 9 channels, regardless of individuals, and hold an accuracy close to the one obtained with all the 64 channels.