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

The identification of psychological stress and disorders like panic attacks and anxiety is vital for improving mental health care. This study investigates the use of electroencephalogram (EEG) signals to predict these conditions, leveraging three deep learning models: a feedforward neural network, a transformer-based model, and a hybrid convolutional neural network (CNN)-long short-term memory network (LSTM). Performance metrics, including accuracy, F1 score, precision, recall, RMSE, and R-squared, were used for evaluation. The feedforward model achieved accuracies of 95.42% and 95.66% for predicting panic and anxiety attacks, respectively, while the transformer-based model also delivered promising results. Notably, the CNN-LSTM hybrid model outperformed the others, achieving 95.90% accuracy for anxiety prediction and excelling in stress level estimation with an RMSE of 0.1671 and an R-squared value of 0.9438. These findings demonstrate the hybrid model's efficacy in predicting psychological conditions and estimating stress levels. This approach holds significant promise for improving mental health assessments, early interventions, and treatment strategies.