EEG Based Mental Arithmetic Task Classification Using a Stacked Long Short Term Memory Network for Brain-Computer Interfacing

Abstract

This paper proposes an electroencephalogram (EEG) based mental arithmetic task classification using a stacked long short-term memory (LSTM) architecture for brain computer interfacing (BCI). EEG signals from 22 channels are taken from 36 subjects, as mentioned in the Physionet database. A deep learning framework based on LSTM is employed to identify and classify the mental arithmetic task by reducing the number of electrodes. Window segmentation is applied for data augmentation as well as feature extraction from all the recorded EEG signals. Eight features per electrode have been fed into the proposed LSTM architecture. The main aspect of this network along with the dropout layers is to enhance feature learning ability and to avoid over fitting.