Novel Attention-Tuned CNN for Emotion and Stress Recognition Using EEG in Cognitive Therapy Sessions

Emotional and stress responses are essential for interpreting cognition and mental health in therapy. Traditional models insufficiently seize the complicated temporal-spatial correlations in EEG data, resulting in imprecise assessments of patients' emotional states. This research presents a Novel Attention-Tuned deep learning (DL) model for recognizing emotions and stress by real-time EEG signals gathered during Cognitive Therapy Sessions, intending to improve recognition accuracy and the interpretability of neural activity connected to psychological states. 2000 EEG data were collected, which undergoes preprocessing employing adaptive median filtering to eliminate noise and artifacts. Discrete Wavelet Transform (DWT) using features is extracted to both temporal and spectral characteristics of the brain’s activity patterns. The See-See Partidge Chicks-driven Attention-Tuned Convolutional Neural Network (SSPC-Att-CNN) model is used to recognize human emotions and stress levels by examining EEG data. This architecture integrates convolutional layers for spatial feature learning with an attention mechanism that adaptively concentrates on the most informative EEG regions corresponding to emotional arousal and stress variations. A SSPC Optimization (SSPCO) algorithm is employed for feature selection, enhancing computational efficacy and robustness. Experimental evaluation was conducted on python 3.10 for standard EEG datasets which demonstrates that the proposed AT-CNN significantly outperforms conventional DL models in recognizing emotional valence in accuracy of 82.10 %, arousal in the accuracy of 97.89%, and stress intensity levels. The system provides real-time cognitive monitoring, giving therapist’s objective insights into emotional regulation and stress patterns. This improves Artificial Intelligence (AI)-assisted psychological assessment tools and promotes a data-driven understanding of emotions in therapy.