Using Simplified EEG-Based Brain Computer Interface and Decision Tree Classifier for Emotions Detection

Abstract

The aim of the paper was to analyze the possibility to recognize human emotions by using a commercially applicable EEG interface and to check how many distinct emotions it is possible to distinguish. The samples were processed to apply to the classifier training. The AutoML software was used to build the decision tree classifier to check the output accuracy and its reliability. Then, we build the classifier for every possible combination. Every EEG band, without distinguishing the high/low frequency, was included in the training or excluded from it. The output of this research was used to determine which EEG bands are the most important in human emotion recognition from EEG data. The AutoML resulted in an accuracy of recognizing four distinct emotions equal to 99.80%. Later, AutoML experiments on each EEG band have shown that the most important specters are respectively, beta, alpha, and gamma, while delta and theta are the less important ones.