Stress Detection Achieves 77.3% Accuracy Using Low-Cost Volatile Organic Compound Sensors

Volatile organic compounds, naturally emitted by the human body, offer a promising new avenue for understanding emotional states, yet this field remains largely unexplored. Nicolai Plintz, Marcus Vetter, and Dirk Ifenthaler, from the University of Mannheim and Technische Hochschule Mannheim, present a comprehensive review and initial feasibility study investigating the potential of these compounds as biomarkers for stress detection. Their work systematically examines existing research, revealing that stress and other affective states demonstrably alter VOC signatures in breath, sweat, skin and urine, although considerable variation exists between individuals. Crucially, the team explores the use of affordable, wearable sensors to capture these changes, bridging a critical gap between laboratory-based analysis and practical, real-world applications, and demonstrating that these sensors can contribute meaningfully to stress detection when combined with physiological data.

Volatile organic compounds (VOCs) represent a novel, yet underexplored, modality for emotion recognition. This paper presents a systematic evidence synthesis and exploratory investigation of VOC-based affective computing using low-cost sensors. A systematic scoping review, analysing 16 studies from an initial pool of 610 records, indicates that stress and affective states are reflected in VOC signatures, specifically aldehydes, ketones, fatty acids, and sulfur compounds, although considerable heterogeneity exists between studies. Current research predominantly relies on laboratory-grade gas chromatography-mass spectrometry or proton-transfer reaction mass spectrometry, whereas wearable sensors currently provide pattern-level outputs without compound-specific identification.

Breath, Skin Response and Heart Rate Analysis

Researchers investigated the potential of using exhaled breath, specifically volatile organic compounds (VOCs), in combination with physiological signals like galvanic skin response and heart rate to detect stress. The study aimed to determine if combining these methods could provide a more accurate understanding of stress responses than single modalities, motivated by the increasing need for non-invasive, real-time stress monitoring. The Trier Social Stress Test, a standardized public speaking and mental arithmetic task, was used to induce stress in participants, alongside the Stroop test as a cognitive stressor. Data collection involved analysing exhaled breath samples for VOCs, recording galvanic skin response and heart rate, and obtaining subjective stress ratings from participants.

The researchers likely used machine learning or statistical methods to analyse the combined data and identify patterns associated with stress. Results suggest that VOCs in exhaled breath can serve as biomarkers of stress, building on previous research showing VOC changes with emotional states and even in relation to diseases. The multimodal approach, combining VOC analysis with physiological measures, offers potential benefits as different modalities capture different aspects of the stress response, potentially improving accuracy. The research points towards the development of wearable sensors and devices for real-time stress monitoring in various settings, including academic institutions and healthcare.

Analyzing VOCs is complex, requiring sophisticated technologies and data processing techniques, and is affected by environmental factors and individual variability. Standardized methods for collecting and analysing breath samples are needed, as is careful interpretation of VOC patterns. This research explores a promising new avenue for stress detection by leveraging chemical signals in breath, combined with established physiological measures, with significant potential benefits for health and well-being.

VOCs Reliably Indicate Human Stress and Emotion

Scientists conducted a comprehensive review of research into volatile organic compounds (VOCs) as indicators of human emotion, analysing data from 610 records and focusing on 16 relevant studies. This systematic review revealed that stress and affective states demonstrably influence VOC signatures present in breath, sweat, skin, and urine, specifically identifying aldehydes, ketones, fatty acids, and sulfur compounds as potential biomarkers. A subsequent study involving 25 participants investigated whether low-cost TVOC sensors, specifically the BME688 and ENS160 models, could detect stress-related VOC patterns when combined with physiological monitoring of heart rate, heart rate variability, and galvanic skin response. Experiments revealed that participants identified as high cardiovascular reactors exhibited elevated TVOC levels during an arithmetic stress test.

The data showed substantial interindividual variability in VOC emission, exceeding 80% coefficient of variation, with coupling patterns influenced by baseline emission levels and temporal lags ranging from 30 to 80 seconds. Utilizing a Random Forest-based multimodal classification approach, the team achieved 77.3% accuracy in stress detection using 5-fold cross-validation, with VOC sensors contributing 24.9% to the overall model performance. Leave-one-subject-out validation yielded 65.3% accuracy, underscoring the importance of individual calibration for accurate emotion recognition. These results demonstrate the potential of integrating low-cost sensors into practical systems for affective computing, while also identifying key challenges related to individual variability and the need for larger sample sizes, specifically at least fifty participants, for robust validation.

VOCs Reveal Human Emotional Signatures

This research presents a comprehensive investigation into the potential of volatile organic compounds (VOCs) as indicators of human emotional states, and explores the feasibility of using low-cost sensors for their detection. A systematic review of existing studies reveals that stress and affective states demonstrably influence VOC signatures found in breath, sweat, skin and urine, specifically involving aldehydes, ketones, fatty acids and sulfur compounds, although considerable variation exists between individuals and studies. This work establishes a foundation for understanding which compounds are most consistently linked to emotional responses and highlights gaps in current knowledge. Researchers conducted an exploratory study demonstrating that inexpensive TVOC sensors, when combined with physiological measurements like heart rate and skin conductance, can detect VOC patterns associated with laboratory-induced stress.

Participants exhibiting a strong cardiovascular response to stress showed increased TVOC levels, and a multimodal classification model achieved moderate accuracy in identifying stress, with VOC sensors contributing significantly to the model’s performance. The team acknowledges substantial variability in VOC emissions between individuals, necessitating individual calibration for accurate readings and emphasizing the need for larger-scale studies to confirm these initial findings. The authors recognise that the exploratory nature of the study, with a limited sample size, requires replication with a larger cohort, at least fifty participants, to validate the observed patterns. Future research should focus on addressing the identified inter-individual variability and developing robust calibration methods to improve the reliability and practicality of VOC-based emotion recognition systems, ultimately paving the way for more sensitive and personalised affective computing technologies.