International Journal of Neural Systems, Vol. 0, No. 0 (0) 1–23 c  World Scientific Publishing Company FEATURE EXTRACTION WITH GMDH-TYPE NEURAL NETWORKS FOR EEG-BASED PERSON IDENTIFICATION Vitaly Schetinin 1* , Livija Jakaite 1 , Ndifreke Nyah 1 , Dusica Novakovic 1 , Wojtek Krzanowski 2 1 School of Computer Science and Technology, University of Bedfordshire, Park Square, Luton, UK vitaly.schetinin@beds.ac.uk 2 College of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter, UK The brain activity observed on EEG electrodes is influenced by volume conduction and functional con- nectivity of a person performing a task. When the task is a biometric test the EEG signals represent the unique “brain print”, which is defined by the functional connectivity that is represented by the interac- tions between electrodes, whilst the conduction components cause trivial correlations. Orthogonalisation using autoregressive modelling minimises the conduction components, and then the residuals are related to features correlated with the functional connectivity. However the orthogonalisation can be unreliable for high-dimensional EEG data. We have found that the dimensionality can be significantly reduced if the baselines required for estimating the residuals can be modelled by using relevant electrodes. In our approach, the required models are learnt by a Group Method of Data Handling (GMDH) algorithm which we have made capable of discovering reliable models from multidimensional EEG data. In our experiments on the EEG-MMI benchmark data which include 109 participants, the proposed method has correctly identified all the subjects and provided a statistically significant (p< 0.01) improvement of the identification accuracy. The experiments have shown that the proposed GMDH method can learn new features from multi-electrode EEG data, which are capable to improve the accuracy of biometric identification. Keywords : Biometrics, multi-electrode EEG, Brain functional connectivity, Volume conduction, Feature extraction, Group Method of Data Handling 1. Introduction Recent advances in neural engineering and human- machine interaction based on the electroencephalo- gram (EEG) are receiving much attention and active development in many application areas. In biomet- ric security applications, EEG signals cannot be re- produced by an intruder or remotely captured with sensors 1, 2 and are extremely difficult to imitate. A person cannot be forced to reproduce a biometric test under stress conditions. 3 The recently developed EEG sensor technologies have significantly improved the usability of EEG headsets, thus making EEG- based technologies user-friendly. 4 EEG-based person identification and recogni- tion methods employ different approaches to extract- ing EEG features that can represent a person’s indi- viduality as a “brain print”. The EEG features are typically represented by frequency spectra. 5–7 When EEG recordings are made from a multi-electrode sys- tem, the features are extracted for each electrode. New promising approaches 8–10 to EEG biomet- rics are based on the “connectome” which reflects in- dividual differences in the brain organisation, known * Corresponding author. 1