Emotion Recognition from EEG Signals During REM Sleep
Alimohammad Mahmoudi, Asghar Zarei
2025 33rd International Conference on Electrical Engineering May 13, 2025 DOI: 10.1109/icee67339.2025.11213928 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Methodological study developing and evaluating a machine-learning classification pipeline on an existing EEG dataset Peer reviewed |
|---|---|
| Key points | The proposed pipeline — bandpass filtering, 20-second windowing, Empirical Mode Decomposition, Sample Entropy features, and SVM/KNN/random forest classification — distinguished positive, negative, and neutral emotions in dream reports from REM-sleep EEG with an average accuracy of 93.83%. The authors argue these findings advance understanding of the neural mechanisms behind dream emotions and may inform EEG-based clinical emotional assessment and intervention. |
Abstract
This paper utilizes EEG signals for automatic emotion recognition that occurs while dreaming in Rapid Eye Movement (REM) sleep. As dream contents could be affected by one's mental health, emotion recognition during REM sleep would give an insight into a person's mental well-being. In this study, an automatic method is introduced for analyzing EEG signals from the Dream Emotion Evaluation Dataset (DEED) to assess brain activity patterns representative of different emotional states (such as positive, negative, and neutral) experienced while dreaming. In this approach, EEG signals were prefiltered using a bandpass filter with cutoff frequencies 1 Hz and 45 Hz and then divided into 20-second windows. Then, every filtered window is decomposed into intrinsic mode functions (IMFs) using Empirical Mode Decomposition (EMD) and subsequently, the Sample Entropy (SampEn) feature is derived from every IMF. Finally, three classification models like support vector machine (SVM), K-nearest neighbor (KNN), and random forest (RF) are utilized for classifying the feature vector. The results show that the proposed method is capable of classifying different emotions in dream reports with an average accuracy of 93.83%. It is therefore believed that these findings expand the understanding of neural mechanisms behind dream emotions and how they can introduce further development into EEG-based approaches for clinical environments dealing with emotional assessment and intervention.