A Computerized Approach for Automatic Human Emotion Recognition Using Sliding Mode Singular Spectrum Analysis

Emotion is a biological process owing to the alteration manifested in the human neurophysiological system, triggered consciously or unconsciously by an external stimulus. Physiological signals such as electroencephalogram (EEG) and electrocardiogram (ECG) measure the neuronal and cardiac activities associated with different emotional states. The complex biological activities result in a highly nonstationary nature of the acquired physiological signals, requiring advanced signal processing and machine learning (ML) techniques to identify and classify hidden patterns. This paper proposes a three-step process for comprehensive analysis and classification of human emotional states. The first step comprises decomposing physiological signals into reconstructed components (RCs) using sliding mode singular spectrum analysis (SM-SSA). In the second step, the discriminatory features such as information potential (IP) and centered correntropy (CEC) were computed from the extracted RCs. Afterward, the extracted features were considered input to various ML classifiers to discriminate human emotional states in the third step. The proposed method was studied over two publicly available databases, DREAMER and AMIGOS, for emotion analysis and achieved the highest classification accuracy of 92.38% in classifying human emotions. The obtained experimental results indicate that the proposed method can identify different human emotional states and yield better performance than existing emotion recognition methods.