Classification of Electroencephalograms during Mathematical Calculations Using Deep Learning

Abstract

Classifying Electroencephalogram(EEG) signals helps in understanding Brain-Computer Interface (BCI). EEG signals are vital in studying how the human mind functions. In this paper, we have used an Arithmetic Calculation dataset consisting of Before Calculation Signals (BCS) and During Calculation Signals (DCS). The dataset consisted of 36 participants. In order to understand the functioning of neurons in the brain, we classified BCS vs DCS. For this classification, we extracted various features such as Mutual Information (MI), Phase Locking Value (PLV), and Entropy namely Permutation entropy, Spectral entropy, Singular value decomposition entropy, Approximate entropy, Sample entropy. The classification of these features was done using RNN-based classifiers such as LSTM, BLSTM, ConvLSTM, and CNN-LSTM. The model achieved an accuracy of 99.72% when entropy was used as a feature and ConvLSTM as a classifier.

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.