In the field of mental health, stress plays a significant role in many conditions, including anxiety, depression, and burnout. That’s why the ability to monitor stress in real time is highly desired as a way to take preventative action. Wearable technology is one possible way to tackle the issue since it can continuously collect passive data on physiological biomarkers that can be used to identify stress as it occurs.
New wearable monitoring technologies have emerged that incorporate biometric sensors like heart rate variability (HRV), galvanic skin response (GSR), and electrodermal activity (EDA), all of which have been shown to have a strong correlation with stress levels. The introduction of Machine Learning (ML) techniques to identify and analyze stress behaviors in real time while monitoring physiological signals from a sensor allows for the provision of individualized interventions through a mobile application accessed by the person wearing the device.
The authors of “Wearable Tech for Mental Health Monitoring Real-Time Stress Detection Using Biometric Sensors” believe that these new developments have significant potential to positively impact the field of mental health. That’s because the majority of existing mental health monitoring systems rely on recurring evaluations through self-reports, interviews, or clinical assessments, which are unable to offer free, continuous, real-time data and are compromised by inherent subjectivity, time commitment (or lack thereof), and human error.
How it Works
Fig. 1. Flow Chart for Wearable Tech System
Data Acquisition and Preprocessing: EDA detects changes in skin conductivity, GSR uses skin perspiration to determine emotional state, and HRV tracks pulse variability. Wearables with Bluetooth are used to collect data, which is then preprocessed before being sent to the app. To reduce the high-frequency noise that is polluting the raw data, a Butterworth low-pass filter is applied before any additional processing. Z-score normalization is used to identify and fix outliers, whereas linear interpolation is used to deal with missing values.
Feature Extraction and Selection: Feature extraction uses the input signals that have been preprocessed to compute statistical, time-domain, and frequency-domain features. Root Mean Square of Successive Differences (RMSSD), Mean RR Interval, and Low-Frequency (LF) to High-Frequency (HF) Ratio are HRV parameters that serve as markers of an autonomic nerve system. The GSR characteristics of Skin Conductance Response (SCR), Mean Skin Conductance Level (SCL), and Response Latency show the reaction of the sympathetic nervous system. Stress-induced changes in skin conductance are recorded using EDA features (Phasic and Tonic Components). The decomposition procedure is then condensed to Principal Component Analysis (PCA) to minimize dimensionality often for computational efficiency while preserving the greatest amount of variance in the data.
Machine Learning-Based Stress Detection: The Random Forest (RF) Classifier is used to identify the stress phases because it can handle the high dimension of physiological data and is resilient against overfitting. RF has a large number of decision trees, each of which is trained using a random feature selection. With a majority-wide vote, decision trees are used to make the final prediction. Validated psychological scales, such as the Perceived Stress Scale (PSS), are used to label physiological datasets with stress levels. The classifier is then trained on these datasets. The number of trees and tree depth are two instances of model hyperparameters that are adjusted using Grid Search Cross-Validation.
Real-Time Stress Monitoring and Alert System: The trained RF model functions as a continuous stress monitoring system and is hosted via a cloud-integrated mobile app. When fresh physiological data is received, feature extraction and categorization take place instantly. A threshold-based decision process that divides stress levels into three groups (low, moderate, and high) based on the classifier's probability outputs comes next. The application sends out a warning recommending relaxation techniques like guided breathing, meditation, or physical activity when the user's stress level exceeds a certain threshold.
Real-Time Feedback and Intervention Strategies: Real-time feedback systems are integrated to assist users in managing stress. When these are experiencing stress, the app recommends particular intervention techniques, such as breathing techniques, guided meditation, or physical activity. The software creates intervention plans based on past stress levels and user compliance. Additionally, the proposed system uses Behavioral Reinforcement Learning (BRL) to optimize the way feedback is administered, continuously adjusting the intervention that is used based on how the user responds to explanation.
System Scalability and Cloud Integration: A cloud infrastructure was integrated for data storage, data processing, and model deployment to grow and support a small number of users. Because data is stored in the cloud, it is safe and enables real-time data syncing across many devices. Anytime accuracy is improved or new functionality is added, it may be easily implemented due to the cloud-based model's ease of updating the machine learning algorithm. It also uses distributed computing to query a lot of data from a lot of users, which enables the proposed system to grow without compromising performance. Extended data monitoring is made feasible by cloud-based storage, which reveals trends in user stress over time. Cloud architecture enhances system interoperability by guaranteeing that model upgrades, interaction data, and updates are accessible and constantly tuned for a better user experience.
Results
The proposed system routinely performs better than two existing systems outlined in earlier papers identified by the numbers 14 and 15:
- M. A. Al-Alim, R. Mubarak, N. M. Salem, and I. Sadek, “A machine-learning approach for stress detection using wearable sensors in free-living environments,” Computers in Biology and Medicine, vol. 179, p. 108918, Jul. 2024, doi: 10.1016/j.compbiomed.2024.108918.,
- B. Venkataramanaiah, K. G. Reddy, S. K. S, K. Ramakrishnan, S. M, and S. Rajesh, “Real Time Mental Health Monitoring System using Machine Learning,” 2024 Second International Conference on Intelligent Cyber Physical Systems and Internet of Things (ICoICI), pp. 806–811, Aug. 2024, doi: 10.1109/icoici62503.2024.10696460.
Specifically, it produces a 92% accuracy rate, 91% precision, 92% recall, and 93% F1-score, all of which indicate better classification ability. Existing systems produce maximum accuracy of 85% and 80%, precision of 87% and and 81%, recall of 83% and 78%, and F1-scores of 85% and 79%, respectively.
Fig. 2. Visual Plotted view of Performance Comparison of Different Models.
Further, the proposed system provides higher levels of accuracy in stress level detection.
Fig 3. Stress Level Detection Accuracy – Proposed System vs Existing System
Conclusion
When compared to the existing system, the results of the proposed wearable stress detection system are competitive. The device can monitor stress levels and offer real-time solutions when necessary due to the usage of biometric sensors. The proposed system appears to be suitable for the widespread delivery of individualized mental health care through a smartphone application.
However, the system has some drawbacks: it depends on properly calibrated biometric sensors, which can differ from person to person and impact data quality; external factors such as temperature and activity level can affect sensor performance; and finally, the system relies on mobile devices for data processing and transmission, which places accessibility restrictions on users without smartphones or internet coverage.
Future research might concentrate on the system's ability to adjust to different environmental circumstances, the creation of fresh methods for accurately calibrating sensors, and expanding the proposed model to include the detection of additional mental health disorders.
Interested in learning more about Wearable Technology and Health Monitoring? The IEEE Xplore Digital Library offers over 37,000 publications on Wearable Technology and over 56,000 publications on Health Monitoring.
Interested in acquiring full-text access to this collection for your entire organization? Request a free demo and trial subscription for your organization.
