EEG-Based Emotion Intensity Recognition using Machine-Learning and CNN-Ensemble Models

Emotion intensity recognition is important for understanding mental states and improving human-computer interaction. Recently, electroencephalogram (EEG)-based spectrogram analysis was explored for emotion classification. However, accurately determining emotion intensity levels remains challenging. In this study, we propose a methodology that combines EEG feature extraction with machine learning classifiers for emotion intensity recognition. EEG signals were collected from 20 participants in a semi-controlled experiment. EEG signals were processed to extract nine time-domain signals, five frequency-domain signals, and spectrograms. These features were applied to support vector machine (SVM), random forest (RF), eXtreme gradient boosting (XGBoost), light gradient boosting machine (LGBM), and hybrid convolutional neural network (CNN)-based models. The performance was evaluated using 10-fold cross-validation. The proposed approach effectively classified emotion intensity, achieving an accuracy, precision, sensitivity, and specificity of 0.996 and a kappa coefficient of 0.994 with the CNN+SVM model. The CNN+RF model, optimized for subject-independent prediction, achieved an accuracy of 0.649 and a kappa coefficient of 0.298. Thus, the proposed method can be extended to real-time emotion monitoring in mental health assessments, stress management, and affective computing applicationsClinical relevance- The proposed framework is clinically relevant in mental health assessment because it enables objective emotion intensity recognition from EEG signals. It can aid in detecting emotional distress, monitoring mood fluctuations in patients with psychiatric disorders, and personalizing therapeutic interventions. Real-time implementation can support early diagnosis and intervention of stress-related disorders, thereby improving patient care and well-being.