An open-access article was published challenging our understanding of stress detection by investigating volatile organic compounds (VOCs), revealing a standalone signal from low-cost sensors but critically reflecting on their current incremental value and the rigorous evaluation methods needed for future biomarker research.
Plintz, N., Vetter, M., & Ifenthaler, D. (2026). Volatile organic compounds for stress detection: A scoping review and exploratory feasibility study with low-cost sensors. IEEE Access. https://doi.org/10.1109/ACCESS.2026.3737607
Volatile organic compounds (VOCs) are an underexplored modality for emotion recognition. This paper presents a systematic evidence synthesis and an exploratory feasibility study with low-cost sensors. Study 1, a PRISMA-ScR-guided scoping review, analyzed 12 studies from 679 records (11 primary studies across breath, sweat, and skin; one urinary reported separately). Stress and affect may be reflected in VOC signatures (aldehydes, ketones, fatty acids, sulfur compounds), but findings are heterogeneous, largely laboratory-based (GC-MS, PTR-MS), while wearable sensors provide pattern-level outputs. Study 2 (n=25) explored whether low-cost TVOC sensors (BME688, ENS160) and physiological monitoring (HR, GSR) detect laboratory-induced stress. Stress induction was confirmed (heart-rate dz = 1.94; self-report Friedman p < .001). Group-level TVOC did not change significantly from baseline to stress (dz = −0.21), and circular-shift permutation analyses found no reliable individual-level physiological–VOC coupling (0 of 24 participants; between-person CV > 80%). Under leave-one-subject-out (LOSO) validation, the VOC channel nevertheless classified stress above chance in isolation (63.6%) yet proved redundant with heart-rate and electrodermal features (66.4% without vs. 65.8% with VOC). Within-sample 5-fold cross-validation yielded an inflated 77.6% due to subject-level leakage. Contributions: (1) mapping of VOC biomarker evidence and technological gaps; (2) subject-independent evidence that low-cost TVOC sensors carry a genuine standalone stress-related signal that does not yet add incremental value over cardiovascular sensing; (3) identification of implementation challenges (inter-individual variability, sensor-level missing data) and a methodological demonstration that within-sample cross-validation inverts modality rankings, underscoring subject-independent evaluation as the appropriate standard. All findings are hypothesis-generating and require replication (n≥50).
