A Study of Deep Learning Models for Identifying and Estimating Psychological Stress and Disorders Us Electroencephalogram Signals

The identification of psychological stress and disorders, specifically panic attacks and anxiety, is paramount for mental health care. Utilizing electroencephalogram (EEG) signals for this purpose is crucial and is gaining significance. In this study, we have proposed and investigated three deep learning models for predicting panic attacks and anxiety attacks, such as a feedforward neural network, a transformer-based model, and a hybrid model combining convolutional neural networks (CNNs) and long short-term memory networks (LSTMs). The models were compared and analyzed on significant performance metrics such as accuracy, F1 score, precision, and recall. Our results suggest that the models were successful in predicting panic and anxiety attacks in terms of accuracy, precision, recall, and F1 score. The Feedforward Model achieved 95.42% accuracy in predicting panic attacks and 95.66% in predicting anxiety attacks. The transformer-based model showed promising results as well. The CNN-LSTM hybrid model performed the best out of the three models, with 95.90% accuracy for anxiety attacks and a better performance in estimating stress levels with an RMSE of 0.1671 and an R-squared value of 0.9438. Our results suggest that the hybrid of convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) performs well in predicting psychological stress and disorders using EEG signals, and it also performs well at estimating stress levels. The exceptional performance in stress level estimation suggests that it has the potential to lead to more accurate and reliable mental-health assessments, which improve early interventions, treatment strategies, and support health policy.