Predicting Stress Levels in Special Needs School Teachers Using DASS-21 and Gradient Boosting Machine

The stress experienced by teachers in Special Needs Schools presents considerable threats to both mental health and job performance, making the need for precise and scalable detection techniques essential. This research presents a multimodal framework for stress classification that combines physiological signals from a wearable IoT device with psychological assessment using the DASS-21. Data were gathered from 48 educators, including heart rate (BPM) and body temperature (°C) as physiological indicators and DASS-21 stress subscale scores as psychological indicators. The target variable was categorized as a binary class (stressed vs. non-stressed) utilizing standardized DASS-21 cut-off scores, with physiological thresholds to improve label dependability. The class distribution was examined to reduce class imbalance bias. The dataset was subjected to preprocessing, which involved normalization and feature selection, and was then divided into a 60:40 train-test split. The evaluation of model generalizability was conducted through 5-fold cross-validation. GBM model was utilized to identify non-linear relationships and interactions between features. Performance assessment involved accuracy, precision, recall, F1-score, specificity, and AUC. The suggested model reached accuracy of 94.12%, 87.5%, and 85.71% for the elementary, junior high, and high school groups, respectively, demonstrating consistently high precision and AUC metrics, reflecting strong discriminative ability. These findings indicate that a clearly defined labeling approach, harmonious feature integration, and strict validation procedure facilitate dependable and consistent stress detection. The suggested framework offers a scalable solution for real-time monitoring of mental health in educational environments. Furthermore, the results explicitly confirm that integrating IoT-based wearable systems with the DASS-21 and the GBM algorithm provides a reliable and scalable approach for stress detection among teachers. This study highlights the practical significance of implementing real-time monitoring systems in educational environments, enabling early identification and intervention for stress management. Such a framework not only improves teachers’ mental well-being but also supports institutional decision-making in developing preventive strategies and sustainable mental health programs.