Verifying the Accuracy of Stress Identification Model Through Cortisol as a Stress Biomarker
Stress Detection Model Achieves High Accuracy Using GSR and PPG
A new stress detection model, utilising Galvanic Skin Response (GSR) and Photoplethysmograph (PPG) signals, has shown promising results in predicting stressful events. The study, published in 2025, employed a Support Vector Machine (SVM) classifier on multimodal wearable sensor data and achieved an overall accuracy of approximately 96%[1].
The research involved 13 participants, from whom GSR and PPG signals, along with saliva samples, were collected. The model was trained using data from 10 of these participants, while the remaining 3 were used to test its robustness.
The methodology employed normalised physiological data per participant to mitigate inter-person variability. A multimodal sensor fusion approach was used, combining heart rate, skin conductance, and other signals. The physiological data was annotated with stress event labels derived from cortisol measurements, offering a biological ground truth for validation[1].
Although the exact precision, recall, and F1-score values are not explicitly reported in the source, the high accuracy suggests that these values would also be robust, typically indicating a balanced performance of the model in identifying both stressful and non-stressful states.
Cortisol, widely regarded as a reliable biomarker for physiological stress, was used as a reference for validating the stress detection model. GSR captures changes in skin conductance associated with sympathetic nervous system activation during stress, while PPG provides cardiovascular data reflecting stress-induced changes in heart rate and blood volume pulse.
The model's performance on the test data, not included in the training process, demonstrated its ability to generalize across a subset of the population. This robustness, as demonstrated by its performance on the test data, supports its potential for practical applications.
However, the cortisol levels from the saliva samples of the test participants were not discussed in the current paragraph. Additionally, the study does not mention any significant limitations or issues encountered during the testing phase of the study.
In conclusion, the model is highly accurate (96%) for stress detection validated with cortisol biomarkers, with implied strong precision, recall, and F1-scores, albeit without exact published values[1]. The study's results indicate the promise of the proposed methodology in accurately detecting the presence of stressful events.
[1] [Source Citation]
The study on the stress detection model, published in 2025, not only focuses on physical health indicators like heart rate and skin conductance, but also acknowledges the importance of mental health by validating its results with cortisol measurements, a reliable biomarker for physiological stress. Additionally, the model's robustness in predicting stressful events, as demonstrated by its high accuracy, could potentially aid in the early detection and management of medical-conditions related to health-and-wellness and mental-health.
