From Self Reports to Sensors in Machine Learning and Multimodal Advances in Chronic Stress Detection

Traditional stress assessment methods rely on self-reports, which are subjective and lack real-time responsiveness. Recent advances in multimodal stress detection address these limits by integrating physiological, behavioral, and environmental data, improving accuracy and robustness. This review synthesizes findings from 50+ recent peer-reviewed studies, highlighting the role of wearable sensors and AI models in high-stress environments. Physiological markers such as Electrodermal Activity (EDA), Heart Rate Variability (HRV), and respiration emerge as reliable indicators, with EDA distinguishing stress from cognitive load in mobile contexts. Machine learning and deep learning methods—including Random Forests, CNNs, LSTMs, and federated learning—report accuracies above 90%. Key challenges remain in data collection, feature extraction, fusion, and interpretability, alongside ethical concerns over privacy and fairness. Finally, a roadmap is proposed toward explainable, privacy-preserving stress monitoring systems tailored for chronic stress management.