Real-Time Neural Network System for Identifying Visual Learners from Raw EEG Data

Presenting a new real-time learner identification system, this study uses EEG signals in its unprocessed form to feed a neural network. Conventional analysis of EEG involves pre-processing signals to get information which is contrary to deep learning algorithm by providing easy extractable information fed into machine’s visual processing. Next, there were LSTM, LSTM-CNN and LSTM-FCNN models which were deployed and experimented to locate high-level characteristics in the EEG signals closely related to learning preferences for visuals. EEG data were obtained from 34 healthy subjects under different conditions but with most focus being given to data obtained while the subjects eyes were closed as this data is relatively pure of external interferences and reflect quite well the cognitive states of a subject. Real-time operations of the deep learning models were carefully designed to manage the different signal lengths for efficient and less computational. Out of all the models, LSTM-CNN was found to be the most effective with the model accuracy of 94%, sensitivity of 80%, specificity of 92% and the model F1 score of 94%. These results support the ability of the system to identify visual learning type from the EEG signals and open a way for using the system in individual learning and cognitive assessment.