A NOVEL METHOD OF EEG-BASED EMOTION RECOGNITION USING NONLINEAR FEATURES VARIABILITY AND DEMPSTER–SHAFER THEORY
These days, emotion recognition has been receiving more attention due to the growth of the brain–computer interfaces (systems) (BCIs). Moreover, estimating emotions is widely used in different aspects such as psychology, neuroscience, entertainment, e-learning, etc. This paper aims to classify emotions through EEG signals. When it comes to emotion recognition, participants’ opinions toward induced emotions are really case-dependent and thus corresponding labels might be imprecise and uncertain. Furthermore, it is acceptable that mixture classifiers lead to higher accuracy (ACE) and lower uncertainty. This paper, introduces new methods, including setting time intervals to process EEG signals, extracting relative values of nonlinear features and classifying them through Dempster–Shafer theory (DST) of evidence method. In this work, we used EEG signals which are taken from a very reliable database and the extracted features are classified by DST in order to reduce uncertainty and consequently achieve better results. First, time windows are determined based on signal complexity. Then, nonlinear features are extracted. Actually, this paper suggests feature variability through time intervals instead of absolute values of features and discriminant features are selected using genetic algorithm (GA). Finally, data is fed in the classification process and different classifiers are combined through DST. 10-fold cross-validation is applied and the results are extracted and compared with some basic classifiers. We managed to achieve high classification performance in terms of emotion recognition [Formula: see text]. Results prove that EEG signals can reflect emotional responses of the brain and the proposed method is effective which gives considerably precise estimation of emotions.