Shahed University

Emotion Recognition Using Chaotic Features And Symbolic Dynamic via Neural Networks

Mahsa Kanani | Mohammad Manthouri | Mehdi Abdolssalehi

Date :  2019/02/28
Publish in :    5th conference on knowledge based engineering and innovation
Link :
Keywords :Electroencephalogram, Emotion, valence, Arousal, Chaotic Features, Symbolic Dynamics, Recurrent Quantification Analysis, Correlation Dimension, Fractal Dimension.

Abstract :
Emotion is one of the most critical aspects of human life which is mostly part of the daily decisions of individuals. The purpose of this study is to examine the brain signals of individuals during the onset of emotion. In this research, the qualitative and quantitative analysis of symbolic dynamics, which is a nonlinear method suitable for studying the behavior of biological signals, has been created between different stages of emotional separation. Symbolic dynamical analysis has proven to be ideal for analyzing complex systems and dynamic time series descriptions. In addition, to extraction of nonlinear and chaotic features (correlation dimension, recurrent quantification analysis (RQA)), used principal component analysis method (PCA) to reduce dimensions of features and then classified 3 emotional classes, LALV, HAHV, and neutral using the multilayer perceptron neural network (MLP with different configurations) and K nearest neighbor (KNN). To evaluate the proposed method, electroencephalogram (EEG) signals from the DEAP database is used. In this database, 32 EEG channels were recorded from 32 people while watching music video clips. The results of this study have shown that the highest average accuracy of classifying 3 emotional classes with PCA as the method of selection/composition of the features for the 5-NN is 85. Also, the accuracy of the multilayered perceptron achieved 79. The accuracy after using the PCA method has also increased by 7 (from 78 to 85).