DeepStress: CNN-GRU-Based Workplace Stress Detection using Facial Expressions and Body Posture

Workplace stress is a significant concern, affecting employees' mental and physical well-being. This research proposes a CNN-GRU-based model for workplace stress detection using facial expressions and body posture. The model integrates Convolutional Neural Networks (CNN) for spatial feature extraction and Gated Recurrent Units (GRU) for temporal sequence modelling to enhance classification performance. A dataset containing labelled facial expression and body posture images was used, with 70 % allocated for training, 15 % for validation, and 15 % for testing. The CNNGRU model outperformed traditional CNN and LSTM approaches, achieving an accuracy of 92.6 %, precision of 90.8 %, recall of 89.7 %, F1-score of 90.2 %, and an AUC of 94.3 %. The verification results of the confusion matrix proved that the model correctly identified 89.7 % stress instances and 91.5 % non-stress instances. The model demonstrated strong discriminatory power based on examination results from the Precision-Recall curve and ROC curve analysis, which produced an AUC value of 94.3 %. The training accuracy, together with loss curves, showed convergence stability, which proved that learning processes could succeed without creating overfitting problems. The research demonstrates the ability of CNN-GRU to detect workplace stress automatically, which creates new possibilities for stress monitoring through mental health assessments.