Experimental Analysis of Emotion Classification Techniques

Existing achievements in the domain of HumanComputer Interaction (HCI) intend to attain a more natural interplay between its involved actors. Automatic and reliable estimations of affective states in particular from physiological signals received much attention lately. From the physiological measures point of view, emotion assessment benefits of pure, unaltered sensations in contrast to facial or vocal measures that can be simulated. In this paper, some physiological measures based classification approaches for assessing the affective state are analyzed in different scenarios. The analysis is performed on the data acquired from Eye-Tracker (ET) sensors, as well as for Heart Rate (HR) and Electro-Dermal Activity (EDA) in visual stimuli based experiments. To this end, a comparison between AdaBoost (AB), K Nearest Neighbors (KNN), Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) is accomplished examining entropy indices as primary features.