Modelling of emotions based on EEG signal

Emotions are important not only in human creativity and intelligence but also in human rational thinking, decision making, curiosity and human interaction. These facts have opened new areas for multidisciplinary research in psychology, neuroscience and affective computing. The electroencephalogram (EEG)-based emotion recognition is an aspect of affective computing (AC) with challenging issues regarding the feature extraction from EEG and learning paradigm to achieve a better classification performance. In this thesis, the conscious processing of audio-visual emotional stimuli is investigated using EEG data. The changes in EEG data and patterns of interactions between eight brain regions correlated to emotions are estimated using various feature extraction methods. The subject-independent patterns are selected and then categorized using various machine learning techniques in a supervised manner. Subsequently, a novel biologically plausible emotion recognition neural network (ERNN) is proposed based on the connectivity features. The proposed EEG-based emotion recognition system comprises six layers; including spectral filtering, a shift register memory, two layers for estimation of coherence between each pair of input signals and a two-layer of radial basis function (RBF) type learning algorithm.Doctor of Philosophy (SCE