Multilayer perceptrons for the classification of brain computer interface data

Fast and simple classification methods for brain computer interfacing (BCI) signals are indispensable for the design of successful BCI applications. This paper presents a computationally simple algorithm to classify BCI data into left and right finger movements of the subjects. A two-class output multilayer perceptron (MLP) performs the classification. Our approach is attractive for providing an optimal combination of 1) computational efficiency, 2) classification accuracy (training: 100% and testing: 64%) and 3) minimal feature extraction (two channels out of a 28-channel EEG trial). The channels selected to be extracted (C3 and C4) not only greatly reduce dimensionality, but also refer to the central parts of the brain that decide left- right cognition, greatly enhancing the classification task. The results obtained are promising, and hold much potential for further investigation.