Emotional State Recognition Based on Brain and Peripheral Signals Using Multi-Class Optimized FSVM Classifiers

Emotions are excellent sources of information in communication, and emotion classification is one of the most sophisticated topics in biomedical signal research. This paper proposes a novel approach based on a fuzzy support vector machine (FSVM) classifier to recognize two of the human’s emotion states (Arousal and Valence), each of which includes positive and negative states, according to Electroencephalography (EEG) signals. The purpose is to attain high accuracy via fuzzification of training data and test data. Data fuzzification not only reduces noise level and data size, but also increases accuracy. In this paper, the KPCA method is used in order to reduce the space dimension due to the greatness of feature space as well as the time consuming nature of SVM training process. In order to obtain the optimal parameters of FSVM core and KPCA parameters - for maximization of the accuracy - for a two-dimensional plan (Valence-Arousal), the Tabu Search Algorithm (TSA) is used. The simulation results demonstrate that the optimized fuzzy classifier improved diagnosis of emotional states by about 3 to 7 percent.

Emotional State Recognition Based on Brain and Peripheral Signals Using Multi-Class Optimized FSVM Classifiers Keywords:

emotion classification , optimization , fuzzy support vector machine (FSVM) , KPCA , Tabu search algorithm (TSA)