LAUSR

The main purpose of this paper is the assessment of emotions using Electroencephalogram (EEG) and peripheral physiological signals and improve recognition accuracy of emotional states using combination mechanism. At the first step, according to the type of signals, effective features were extracted in the time and frequency domains; then, by using the Fisher’s Linear Discriminant (FLD) method, the most effective features were selected. Based on these features, six classifiers were used: Support Vector Machine (SVM), Nearest Mean (NM), K-Nearest Neighborhood (K-NN), 1-Nearest Neighborhood (1-NN), FLD and Linear Discriminant Analysis (LDA). They classified emotions in two classes (low and high) through arousal, valence and liking dimensions. The leave-one-out cross-validation (LOOCV) method has been implemented to evaluate the performance of classifiers. To enhance the accuracy of classification, combination at feature and classifier levels were performed. Via the concatenation method, combination at feature level was done. Then, by Majority voting, Fixed and Stacking algorithms, combination at classifier level were implemented. Results showed that these classifiers were selected properly and comparing with previous works, good improvements were achieved by them. Finally, by using combination methods, obtained recognition accuracy was much more reliable and combination at classifier level resulted in significant improvement