Dynamic Emotion Recognition from EEG Signals: A Hybrid Approach Combining DNN and RNN Architectures

This research introduces an advanced deep learning architecture capable of accurately detecting stress and identifying brain signals related to stress relief by analyzing EEG data. Current methods for stress detection often struggle to precisely distinguish between various stress patterns and do not fully leverage advanced machine learning techniques for accurate classification of stress levels based on EEG data. The study focuses on differentiating between two essential categories of stress: ontological positive stress, or eustress, and negative stress, which is referred to as distress, resulting in suffering and adverse mental conditions. By integrating both Deep Neural Networks (DNNs) and Recurrent Neural Networks (RNNs), this hybrid approach surpasses traditional analysis techniques. The high level of integration achieved enables sophisticated signal processing of time-frequency domain features extracted from the initial EEG signals. The significant approach called the Matthew Correlation Coefficient Advanced (MCCA) feature extraction technique, is much more accurate in distinguishing stress patterns compared to other methods. The proposed DNN-RNN hybrid model also performs with highly encouraging results standards, with a targeted classification accuracy rate of 94.19%. The model also shows high efficiency for each assessment parameter, such as specificity, sensitiveness, and F1 score and recall rate, all over 89.00%. Combining deep learning with neurophysiologic data reveals new possibilities for individualized approaches to mental health and stress-coping strategies.