Clinical Report: Utilizing Deep Learning Techniques for Recognizing Stress-Related Human Activities
Overview
This study demonstrates the efficacy of deep learning architectures, particularly Transformer models, in recognizing stress-related human activities using sensor-generated time-series data. The results indicate significant improvements in classification accuracy, suggesting potential applications in mental health monitoring.
Background
Human Activity Recognition (HAR) is increasingly important for assessing mental and physical health, particularly in the context of rising stress-related disorders. With 75% of individuals reporting stress in their daily lives, effective monitoring tools are essential for timely interventions. Leveraging deep learning techniques can enhance the detection of stress-related behaviors through wearable technology.
Data Highlights
Model
Classification Accuracy
Transformer
97.83%
LSTM
97.36%
RNN
92.4%
Key Findings
Transformer models achieved the highest classification accuracy of 97.83% for stress-related activities.
LSTM and RNN models showed lower accuracies of 97.36% and 92.4%, respectively.
Window size and overlap ratio significantly impact classification performance.
The proposed approach outperformed previous models on the Stressense dataset.
HAR can effectively monitor stress-related behaviors through non-intrusive wearable devices.
Clinical Implications
The findings support the integration of deep learning models in wearable devices for real-time stress monitoring. Clinicians can utilize these technologies to enhance patient assessments and interventions for stress-related disorders.
Conclusion
The study underscores the potential of advanced deep learning techniques in improving the accuracy of stress detection through HAR, paving the way for innovative mental health monitoring solutions.
