Wearable Stress Classifiers Face "Structural Ambiguity" Challenges

Wearable devices designed to classify stress often show impressive average performance, but this can hide instances where they completely fail for specific individuals. This failure can occur due to what researchers term "structural ambiguity," where an individual's physiological signals, while plausible on their own, exhibit inter-signal patterns that deviate significantly from their non-stress baseline. To address this, the study introduces the Individual Conformal Coupling Monitor (ICCM). This lightweight and transparent pre-inference tool quantifies subject-specific coupling divergence. It can then route data windows to either be classified, deferred for further review, or abstained from classification, all without needing to retrain the main stress classifier. While ICCM showed negative correlations between ambiguity and accuracy across two datasets, further robustness analyses indicated that these correlations were not always statistically significant, and the effect could disappear when outlier subjects were removed. Although ICCM did slightly reduce false positives and withheld some stress windows for a failing subject, it did not fully resolve the missed-stress issue. The research positions ICCM as a valuable, interpretable indicator of unsupported physiology and individual failure, rather than a standalone safety guarantee.