Affective detection based on an imbalanced fuzzy support vector machine

The interpretation of physiological signals is an important subject in affective computing. In this paper, we report an experiment to collect affective galvanic skin response signals (GRS), and describe a new imbalanced fuzzy support vector machine (IBFSVM) for their classification. IBFSVM introduces denoising factors and class compensation factors, thus defining a new fuzzy membership. The effectiveness of IBFSVM is verified on various real and artificial datasets. We define an appropriate evaluation criterion (g_mean) that combines the classification accuracy of positive and negative samples, and show that IBFSVM outperforms traditional support vector machines on imbalanced datasets. By running the IBFSVM for the datasets in our experiment, we can find that the g_mean of happiness, sadness, angry and fear is 85.17%, 86.6%, 87.4%, and 81.53% respectively. So IBFSVM is an effective and feasible solution for imbalanced learning in our experiment.