Three-way Analysis of Multichannel EEG Data Using the PARAFAC and Tucker Models

Changes of hidden sources of the neural electrical activity of a brain over time, as represented by a continuously recorded multichannel electroencephalogram (EEG) at the scalp, can be detected by tensor or multiway decomposition of the EEG records. In this study, the performance of (i) the constrained Tucker model and (ii) the parallel factor analysis (PARAFAC) models are compared on real EEG data.