Invention Description
Accurately detecting stress using wearable devices is challenging due to significant differences in how individuals respond physiologically to stress. Variations in factors such as heart rate, skin conductance, and environmental conditions can reduce the reliability of stress detection models across different users. Many existing systems are trained on limited datasets and fail to generalize well to new populations, especially in complex cases such as individuals with opioid use disorder. This creates a need for more robust approaches that can deliver consistent performance across diverse users and settings.
Researchers at Arizona State University have developed HHISS, a domain generalization framework designed to improve stress detection by using person-wise sub-network pruning and continuous label training. By focusing on shared patterns of stress responses rather than individual differences, HHISS enhances model generalization across diverse populations and environments. The system is particularly effective in challenging use cases, including monitoring stress in individuals with opioid use disorder. This approach enables more reliable and scalable stress detection using wearable devices in real-world conditions. HHISS is an advanced stress detection system that overcomes individual differences to deliver consistent and accurate stress monitoring using wearable devices.
HHISS is an advanced stress detection system that overcomes individual differences to deliver consistent and accurate stress monitoring using wearable devices.
Potential Applications
-
Wearable health devices for stress monitoring for clinical and consumer use
-
Real-world mental health and rehabilitation support tools for opioid use disorder patients
-
Mental health and wellness applications
-
Personalized healthcare and remote patient monitoring
-
Mobile health applications requiring reliable, scalable stress sensing
-
Research tools for studying stress responses across diverse populations
Benefits and Advantages
-
Enhanced model accuracy across individuals and environments by eliminating person-specific variability
-
Prevention of overfitting via person-wise sub-network pruning
-
Robust generalization to unseen stressors and environments using continuous label training
-
Improved scalability to various wearable platforms
-
Effective use of continuous stress labels for precise model training
-
Superior performance over existing state-of-the-art methods as validated on multiple datasets
-
Proven scalability and feasibility for mobile and real-world applications
For more information about this opportunity, please see