A Cyber-Physical Fusion System for Stress Detection Using Multimodal and Social Media Data

Stress is identified as one of the most common human responses to physical, mental or emotional pressure. Long-term stress can cause cardiovascular diseases, depression, anxiety and even death. Stress can be recognized by observing physiological activity data and social media posts of individuals. This explorative study is performed to find the effect of fusion of physiological measurement with social media textual posts for classifying stress. The proposed model implements Heart Rate Variability (HRV) datasets as physiological stress datasets and social media post dataset as textual dataset. At first the datasets were individually implemented with different machine learning models to find the best fit model. It is shown that Random Forest showed the best classification result with an accuracy of 99.85% for the HRV data and the Logistic Regression model performed best for the social media data with an accuracy of 96.4%. The two models are combined using fuzzy fusion technique with an accuracy of 98%. To our knowledge, the fuzzy fusion technique for combining physiological and textual data is a novel approach for stress detection with significant applicability.