Improving Physiological Emotion Recognition Using Hybrid SVM-Firefly and Random Forest Models

Abstract

Aiming at the problems of low recognition accuracy of physiological and emotional signals and susceptibility to interference during the recognition process, an improved algorithm based on machine learning is proposed. The support vector machine algorithm and firefly algorithm model are used for data classification and recognition of physiological signals. The random forest algorithm is added to improve the recognition and decision-making performance of the algorithm model. The experiment is based on the publicly available DEAP dataset, which includes 32 subjects, 1,280 3-minute multi-modal physiological models, and a sampling rate of 360Hz. Taking five emotions of happiness, sadness, anger, fear, and neutrality as tasks, the accuracy and F1 value are used for evaluation, and comparisons are made with support vector machine, random forest algorithm, convolutional neural network, and long short-term memory network fusion algorithm. The results showed that among the five types of emotion changes, sample 2 had the highest recognition accuracy for different emotion signals, with an overall accuracy of over 90.00%, indicating that this model could effectively extract features from the signals. The highest accuracy of the proposed model reached 94.60%, which was approximately 3.10% higher than that of the sample with the lowest accuracy. Its average F1-macro reached as high as 90.3%, outperforming the comparison model in some emotion categories. However, the lowest F1 value of 84.9% occurred in emotion state 4, slightly lower than that of the comparison model. The designed model has better accuracy and higher recognition performance in emotion recognition. This has good guiding value for the research on improving the accuracy of emotion recognition for different signals at present.