Neural network-based predictions of hand movements using simulated and real EEG data

Abstract

Recently, neural networks have been used as a tool for the classification of spatio-temporal EEG patterns arising from a hand movement experiment. Results indicated that, based on single-trial EEG data recorded before movement, the side of hand movement can be predicted with fairly high precision, but variability of results raised the question of their validity. In order to validate results obtained with real EEG data, an equivalent simulated movement experiment was performed. Alpha band rhythms composed of two components were simulated as a superposition of two second-order autoregressive (AR) processes. These simulated EEG data were then filtered, and their power values calculated and used as features in a classification task. Systematic analysis of the sensitivity of the classification results on various simulation parameters was performed. The analysis showed that the Cascade-correlation (CC) network is able to perform satisfactorily in an extremely noisy environment.