A feature-based dual-layer ensemble classification method for emotional state recognition / Mehdi Malekzadeh

Mehdi , Malekzadeh (2017) A feature-based dual-layer ensemble classification method for emotional state recognition / Mehdi Malekzadeh. PhD thesis, University of Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (1659Kb)
Preview	PDF (Thesis PhD ) Download (2757Kb) | Preview

Abstract

It is important to recognize an individual’s emotional state as it can be used in many disciplines and research in areas such as medicine and education. Human emotions can be recognized through the analysis of several modalities, which include speech, facial appearance, gestures, and human physiology. Among the different modalities of human emotion expression, the physiological data that can be gathered from people, especially the speech impaired people is probably the most reliable for human emotion recognition. The physiological modality has the advantage of being more robust against possible artifacts of human interpersonal hiding since they will be instantaneously managed by the human autonomic nervous system. The current automatic physiological-based emotion recognition systems call for improvement in two main respects which are applying a feature selection method for selecting an optimal feature subset and selecting a suitable classifier that maximizes the classification performance of the emotion recognition system. The main aim of this research is to improve the classification accuracy of physiological-based emotion recognition systems by proposing a feature-based dual-layer ensemble classification method. In addition, we analyse the accuracies of various classification methods with different physiological modalities and feature selection methods in order to understand the effect of each component on the overall performance of the emotion recognition system and recommend a system's design that can achieve the best classification accuracy for emotion recognition systems. The results show that for single classifiers, Support Vector Machine (SVM) achieved the best classification method to be used for developing emotion recognition system and there is no single type of modality that is suitable for all the classifiers. In addition, feature selection methods have positively contributed to the improvement of multi-classifier methods compared to single classifiers. Compared to the best single classifiers, the proposed feature-based dual-layer ensemble classification method has improved the accuracy around 5% to 17%. The proposed classification method can be used or tested on other emotion databases or even on other medical diagnosis problems that use physiological data.

Actions (For repository staff only : Login required)