Evaluation of Galvanic Skin Response (GSR) Signals Features for Emotion Recognition

Over the years, physiological signals have shown its efficiency in emotion recognition. Galvanic skin response (GSR) is a quantifiable physiological signal generated from the change of skin conductance in response to emotional stimulation. Understanding human emotions through GSR signals can be a challenging task because of the characteristic’s complexity. The current performance on the analysis of GSR signals has yet to be satisfactory due to a lack of detailed evaluation on the performance of features extracted from GSR signals. Previous studies have compared the recognition rates between different physiological signals between electroencephalogram (EEG), electrocardiogram (ECG), and GSR as a group or focused on the performance of emotion recognition using a fusion of signals. This paper presents an evaluation of extracted features specifically from GSR signals from a public dataset named as AMIGOS database. The MATLAB software was used for the simulation. In the study, feature extraction techniques were performed to extract features in time domain and frequency domain features. These features are ranked using the one-way ANOVA method in MATLAB. Several subsets of different number of features based on the type of feature and significance level were formed for optimum selection. The state of art classification algorithm for GSR which is Support Vector Machine (SVM) was employed to evaluate the classification performance using the ranked features. The methodology proposed by this study was able to achieve high accuracy rates that are comparable with existing studies that had employed the same AMIGOS database. The frequency domain features achieved the highest accuracy for all four emotion classes.