AI-Driven Multimodal Stress Monitoring in the Workplace Using Wearable Physiological Biomarkers

Chronic stress adversely affects professional performance and long-term mental health, necessitating continuous, objective monitoring solutions beyond traditional clinical assessments. This study presents an AIdriven framework for multimodal stress prediction that integrates wearable physiological sensors with validated psychological assessments, advancing personalized and remote mental health monitoring. We monitored 64 experienced STEM professionals (≥4 years industry experience) longitudinally across three phases (Baseline, Residency, Post-Residency) over one year. Our framework fuses continuous physiological biomarkers-electrodermal activity (EDA), heart rate variability (HRV), respiratory rate (RR), and skin temperature-captured via Empatica EmbracePlus devices, with weekly Perceived Stress Scale (PSS) scores to enable three-class stress categorization (Low/Moderate/High). Although ensemble methods achieved 96% accuracy, they exhibited overfitting and limited generalizability. A deep neural network with adaptive class weighting offered stronger clinical relevance, achieving 87% sensitivity for Moderate Stress detection and supporting early intervention. A lightweight logistic regression model (89% accuracy) further enables deployment in resource-constrained environments. This research establishes the first validated framework for longitudinal stress detection in real-world professional settings using multimodal wearable data. The approach enables integration into digital health ecosystems, supporting both personalized stress management and scalable remote monitoring for preventive mental healthcare.