Cross-subject emotion recognition using hierarchical feature optimization and SVM with multi-kernel collaboration.

Due to individual differences, it is greatly challenging to realize the cross-subject multiple types of emotion identification. In this research, a hierarchical feature optimization method is proposed in order to represent the emotion states based on peripheral physiological signals effectively. Firstly, sparse learning combined with binary search is employed to achieve feature selection of single signal. Then an improved fast correlation-based filter (IFCBF) is proposed to implement fusion optimization of multi-channel signal features. Aiming at the limitation of support vector machine (SVM) using single kernel function to make decision, multi-kernel function collaboration strategy is proposed to improve the classification performance of SVM. The effectiveness of the proposed method is verified on DEAP dataset. Experimental results show that the proposed method of hierarchical feature optimization and SVM with multi-kernel function collaboration presents competitive performance for cross-subject multiple types of emotion identification with accuracy 84% (group 1) and 85.07% (group 2).