a real-time electroencephalography classification in emotion assessment based on synthetic statistical-frequency feature extraction and feature selection

purpose: to assess three main emotions (happy, sad and calm) by various classifiers, using appropriate feature extraction and feature selection. materials and methods: in this study a combination of power spectral density and a series of statistical features are proposed as statistical-frequency features. next, a feature selection method from pattern recognition (pr) tools is presented to extract major features and apply to classifiers. results: the experimental results on various classifiers demonstrated the priority of proposed emotion assessment system to the previous ones where back-propagation neural network was the most accurate classifier to complete the proposed system and linear discriminant analysis was the best choice regarding to the accuracy and runtime of the system. conclusion: in this paper we proposed a prominent method that led to a highly accurate system with three emotion states. in this regard, unequal numbers of experiments on different emotion states were employed. this idea indicated that in order to avoid domination of one emotion state rather than other states in self-induced emotion signals unequal number of different states should be applied.