An optimized multi-label TSK fuzzy system for emotion recognition of multimodal physiological signals^*

Nowadays, as intelligent detection devices become reachable, how to integrate multimodal physiological signals to identify human emotions becomes a hot topic. However, the corresponding relationship between physiological signals and emotions is often difficult to be modeled mathematically, and too many signals will bring redundant features and reduce the accuracy and efficiency of emotion classification. In order to solve these problems, this paper adopts multi-label TSK fuzzy system to establish fuzzy rules for emotion recognition. Meanwhile, this paper proposes an optimization method of fusing subspace clustering and fuzzy C-means clustering to solve the problem of establishing fuzzy rules for high-dimensional features. We also consider uniform regularization to balance the trigger strength of different fuzzy rules. A dataset named DEAP is used to test the accuracy of our method. The results showed that Multi-label TSK fuzzy system with subspace fuzzy C-means and uniform regularization has a nearly 10% improvement in accuracy compared with the traditional SVM, and it’s also helpful to explore the correlation of different labels such as arousal and valence, which can contribute to the further study of affective computing.