EEG-Based Communication: A Time Series Prediction Approach

Recently, a new technology known as the braincomputer interface (BCI) has received a substantial amount of interest among various research groups worldwide. The human brain can be represented by self-organising and complex biochemical states. Due to continuous neuronal activity in the brain, chaotic electric potential waves are observed in Electroencephalogram (EEG) recordings of the brain. A BCI involves extracting information from the highly complex EEG. This is achieved by obtaining the dominant discriminating features from different EEG signals recorded during specific thought processes. A class of features is usually obtained from each thought process and subsequently a classifier is trained to learn which feature belongs to which class. This ultimately leads to a system that can determine which thoughts belong to which set of EEG signals. This work outlines a novel method which utilises cybernetic intelligence in the form of Neural Networks (NN). Three NNs are coalesced to perform simplified simulations of a number of the characteristic and complex processes that are sub-consciously performed in the human brain. These include prediction, feature extraction and classification. These processes are combined in this system to produce a pattern recognition system which distinguishes between similar complex patterns from a noisy environment with classification accuracy which compares satisfactorily to current reported results. The classification accuracy is achieved by increasing the separability between the features extracted from two EEG signals recorded from subjects during imagination of left and right arm movement.