Epilepsy is the most common chronical neurological disorder. Nearly 1% of the world population suffer from recurring epileptic seizures and many of these patients would benefit from a seizure warning device. However, the problem of anticipating epileptic seizures with sufficient specificity and sensitivity necessary to realize a reliable warning system for these patients remains unsolved for nearly two decades of investigations by various authors from multiple fields of research. Recent investigations [1], [2] indicate that multivariate algorithms could be helpful to the task of anticipating epileptic seizures. Even though the coupling of neighboring EEG channels has been studied in a signal prediction approach [5], channels of different brain regions have not been considered in CNN signal prediction approaches up to now. In this paper we propose statistical analysis of EEG recordings based on signal prediction approach, leading to the identification of regions of the brain that are spatially separated but seem to interact prior to a seizure.