Frequent, sustained stress is linked to poor health and requires monitoring for early intervention. The electrocardiogram (ECG) is a promising biomarker because it can be recorded non-invasively and continuously using wearable devices. However, tracking stress with ECG is challenging because daily activities elicit responses similar to mental stress, and the variety of mental stimuli individuals encounter complicates the use of machine learning (ML) models trained on a limited set of stressors.

In this study, Bülent Ündes and colleagues evaluate machine learning-based mental stress detection using 127 participants across multiple conditions, testing (1) discrimination against a composite no-stress baseline, (2) generalisation to unseen stressors and participants, and (3) robustness under reduced sampling rates and number of features for wearable settings.

Findings
The results demonstrate that ML models can detect mental stress with high sensitivity and remain robust to lower sampling rates and fewer features. Generalisation to novel stressors was stressor dependent. Importantly, these results highlight challenges in distinguishing stress-related cardiac responses from those caused by physical exertion, revealing critical limitations of single-sensor ECG approaches for mental stress detection.

Conclusion
Single-sensor ECG ML models are insufficient for reliable mental stress monitoring in real-world settings where physical activity occurs. Multi-sensor approaches or contextual information is necessary before clinical or consumer deployment.