EEG based Directional Signal Classification using RNN Variants

Abstract

EEG(Electroencephalogram) signals generated within the brain can be extracted using sensors. Thus generated signals can be classified based on the feature that are embedded within it. The signals once recognized can act as alternative inputs for users suffering from severe motor impairment. The inputs can be used for motion signal i.e directions left, right, up and down. In this paper, the raw EEG signals and power signals generated from NeuroSky Mindwave device have been classified using deep neural networks. Bi-directional Long Short Term Network architecture(Bi-LSTM) and a model which uses Long Short Term Memory(LSTM) with Attention layer have been implemented for the purpose. An accuracy of 56% was obtained using bi-directional LSTM network with raw signals, 44.75% accuracy with power signals, and with attention network using raw signals an accuracy of 63% was obtained.