Deep learning-based intelligent assessment and early warning system for employee mental health status

Employee mental health is a critical component of organizational productivity and sustainability. With increasing work pressure and digital overload, early identification of mental health issues is essential for timely intervention. However, existing methods often rely on static surveys or manual evaluations, which lack real-time analysis, contextual understanding, and predictive capability. These conventional approaches are limited by low accuracy, delayed response, and poor adaptability to dynamic behavioral changes. To address these limitations, this paper proposes a Deep Learning-based Intelligent Assessment and Early Warning System that leverages a Bidirectional Long Short-Term Memory (BiLSTM) + attention mechanism. The BiLSTM model captures sequential patterns and contextual cues from multimodal data, including textual sentiment from communication logs, physiological indicators, and behavioral metrics, to accurately evaluate and predict an employee's mental health status. The proposed system is capable of continuously monitoring employees, identifying early warning signs of stress, anxiety, or burnout, and notifying the support system while ensuring privacy. The system achieves 89.3% accuracy, a macro F1-score of 91.3%, a Mean Absolute Error (MAE) of 90.1%, and a ROC-AUC score of 94.5%. It provides a significantly improved early warning lead time of 5.6 days compared to traditional models. These findings confirm the viability of the proposed system for ethical, real-time deployment in corporate mental wellness programs.