Dynamic entropy-based pattern learning to identify emotions from EEG signals across individuals

Emotion plays an important role in mental and physical health, decision-making, and social communication. An accurate detection of human emotions is critical to ensure effective interaction and activate proper emotional feedback. In the existing emotion recognition methods, poor generalization capability caused by individual differences in emotion experiences is still a problem. This article proposes a new framework of dynamic entropy-based pattern learning to enable subject-independent emotion recognition from electroencephalogram (EEG) signals with good generalization. Firstly, we exploit dynamic entropy measures in quantitative EEG measurement to extract consecutive entropy values from EEG signals over time. Then, based on the concatenation of consecutive entropy values to form feature vectors, the dynamic entropy-based patterning learning can be able to achieve subject-independent emotion recognition across individuals to obtain excellent identification accuracy. Experiment results show that the best average accuracy of 85.11% is reached to identify the negative and positive emotions. Besides, by comparison with the recent researches, the results have fully demonstrated that our method can achieve excellent performance for emotion recognition across individuals. In summary, an universal and subject-independent emotion recognition method with excellent generalization capability is developed by the proposed dynamic entropy-based pattern learning, which may have the great application potential to address the emotion detection in healthcare decision-making and human-computer interaction systems.