Experimental design of emotion recognition based on machine learning

In this paper, a wrist bracelet was used to measure the subjects' heart rate signals and skin electrical signals, and three types of emotions (positive, neutral and negative) were induced by designing an experimental environment. The heart rate and skin electrical signals corresponding to emotions were collected through the bracelet, and the recognition accuracy of the three types of emotions was finally obtained through data preprocessing and the support vector machine model in machine learning. At present, how to build an experimental environment that can fully induce subjects' emotions is a major difficulty in emotion recognition, so this paper provides a specific emotion recognition environment design and detailed steps. Meanwhile, through a large number of experiments, the kernel function in the support vector machine model, namely Gaussian kernel function, is determined, and grid search is introduced to search for hyperparameter C. To get the optimal parameters and results. A total of 20 people were measured in this experiment. The experimental results obtained by SVM model were positive: the recognition accuracy of skin electrical signal was 0.8422, the recognition accuracy of heart rate signal was 0.8345, and the recognition accuracy of skin electrical signal and heart rate signal feature fusion was 0.8832. Negative: the recognition accuracy of skin electrical signal is 0.9812, the recognition accuracy of heart rate signal is 0.9385, and the recognition accuracy of skin electrical signal and heart rate signal feature fusion is 0.9902. Neutral: the recognition accuracy of skin electrical signal was 0.6403, the recognition accuracy of heart rate signal was 0.5308, and the recognition accuracy of skin electrical signal and heart rate signal feature fusion was 0.5438