EEG based mental task classification using arithmetic operations

Abstract

The human brain is exposed to a variety of positive and negative emotions, such as stress, happiness, frustration, and attention, while completing any task. Techniques for brain-computer interface (BCI) can be used to study and identify these emotions. The most widely used neuroimaging technique for the examination of brain activity is the Electroencephalogram (EEG), which is affordable, reliable, and non-invasive. Here, the EEG signal pattern during the introduction of a mental task is investigated. In the study, the dataset was taken from PHYSIONET, named as EEG during mental arithmetic tasks. Data of ten healthy volunteers was taken among 36 participants in two different modes-rest and mental task-with a sample frequency of 500 Hz. Overall, all, 11 frequency domain features were examined in this work, including power band features, band power ratios, and relative powers. Additionally, two wavelet features-spectral entropy and Shannon entropy-were added to the feature set. From time domain features, mean, standard deviation and variation features were also extracted. Six Classifiers were used and among them neural networks provided highest accuracy of 98.8%.