Human emotion classification enabled by EEG signal analysis and machine learning

In recent years, automated human emotion recognition system poses numerous challenges with large and complex data and its vast computation. In this work, we consider a three-dimensional continuous valence-arousal-dominance framework for data representation in space using the DEAP dataset of 32 participants. An experiment was carried out to test practicality of the proposed system using the EMOTIV Insight headset electroencephalogram (EEG) data from four channels. We utilise a tunable-Q wavelet transform (TQWT) algorithm to extract frequency domain characteristics of the signals and classify using time windows of two seconds. The extracted power of the signals was identified as features from different frequency bands and the K-nearest neighbour (KNN) and random forest (RF) classifiers resulted in an accuracy of 97.8%.