Recurrence Quantification Analysis and Neural Networks for Emotional EEG Classification

Abstract

Purpose: There are many benefits for emotion recognition and classification in human-computer interaction, social communications, gaming industries, and entertainment. Therefore, physiological responses of emotions have been receiving a significant attention. However, the assumption of nonlinear characteristics of physiological signals is usually disregarded. Basic methods: In the current study, a novel approach for classification of emotional states is presented using electroencephalogram (EEG) signals and nonlinear methodology. Applying 3 channels of EEG from eNTERFACE06_EMOBRAIN database, some measures of recurrence quantification analysis (RQA) (including: recurrence rate, deterministic, average line length of diagonal lines, entropy, laminarity, and trapping time) are calculated in 3 emotional states (exciting negative, neutral and exciting positive). These features are considered as inputs of the multilayer perceptron, time delay neural network, and probabilistic neural network (PNN) classifiers. Main results: Based on the RQA measures and PNN, the emotion detection system outlined here is potentially capable of classifying 3 emotional categories. The accuracy rate of 99.96% is attained which is comparable to the results achieved by others.Conclusions: The results show that the proposed methodology can be used as an appropriate tool for emotion recognition.