Classtfication and Mathematical Modeling of Human Emotions from EEG Signal

Abstract

Cognitive state estimation shows the subjective mental changes with the environmental constraints which can be used for diagnosis of cognitive behavior. A cognitive model will support and facilitate the development of affective systems in emotion studies and act as a unifying platform in physiological research area. In recent years, there has been an increasing interest in applying techniques from the domains of nonlinear analysis in studying the mental behavior of a dynamical system from an experimental time series such as EEG signals. A lot of research has been carried out to study on human brain response while the subject is in relax or performing different mental task with sustained attention or listening to different kinds of music, as well as different emotion related activity. High frequency component and low frequency component contained in a brain signal with different mental activity is proven as a cognitive factor to human emotion recognition system and can be shown through the variations of human brain signal. Electroencephalographic (EEG) technology has enabled effective measurement of human brain activity, as functional and physiological changes within the brain may be registered by EEG signals from the variations of alpha, beta, delta, theta frequency bands. The EEG signals are collected from several healthy adult subjects and processed using signal processing algorithms in C/C++ source code and MATLAB to extract the effective features to classify the emotional states through the spatial and temporal analysis, discrete wavelet transform, fast Fourier transform etc. Useful information is extracted from the processing of EEG signal, and different machine learning algorithm are used to identify the different brain response from the signals to classify the emotional states using multiclass support vector machine (MCSVM). The classification of different emotions is validated using artificial intelligent techniques, i.e. neural network. The recognition of human emotion plays a vital role in physiological research area but in case of real-time application and practical hardware implementation of human emotion based systems a mathematical background of emotions is really needed. Mathematical modeling of emotional states plays a significant role in this scope which can correlate between human cognition, emotion and mental behavior. In this work, new approach is proposed to model the emotional states with mathematical expressions based on wavelet analysis and trust region 4- algorithm for the non-linearity and non-stationarity of EEG signal. Daubechies4 wavelet function ("db4") is applied on different recognized emotional states such as relax, memory, pleasant, fear, motor action (MA), enjoying music (EM) to extract the wavelet coefficients of these different states. The emotional states are modeled with different mathematical VI LI expressions. The brain signals are composed of composite frequency components. So, the proposed model of the emotional states will be the sum of the sinusoidal functions consisting the composite frequency components. To model the emotional states the coefficients can be obtained by trust-region algorithm for non-linear EEG data which can be verified with these subband wavelet coefficients. The adjusted R- square percentage and the sum of square error will optimize the performance of proposed model. The higher rate of adjusted R-square percentage and lower percentage of SSE and RMSE will validate the developed cognitive model. To propose a proper mathematical model of the brain signal of different emotional states proper effective channel is needed to select in order to reduce the feature size without any performance degradation. In this work a way is develop to propose the effective channel for emotion classification based on temporal and spectral analysis. The performance of the proper selected channel is more robust to classify and model the effective emotional states.