Physical, Social and Cognitive Stressor Identification using Electrocardiography-derived Features and Machine Learning from a Wearable Device

Anxiety is a prevalent and detrimental mental health condition affecting young adults, particularly in college students who face a range of stressors including academic pressures, interpersonal relationships, and financial concerns. The ability to predict anxiety would help create individualized treatment. There is a need for objective and non-invasive continuous monitoring tools that allow for the prediction of anxiety. However, the generalizability of physiological changes across various stressors and participants must first be examined. The aim of this work is to examine the relationship of different stressors on heart rate variability in combination with machine learning models to assess binary and multi-class classification performance using electrocardiography derived features from a wearable device. Twenty-six college students performed a series of non-stressful and stressful conditions while wearing a Hexoskin smartshirt. The performance of binary and multi-class ML classifiers of stressor type was evaluated. Condition-wise binary classification accuracy of 76.2% and multi-class classification accuracy of 79.1% were achieved using a support vector machine (SVM) architecture. These results contribute to our understanding of individual anxiety symptom detection using ML and offer implications for applying similar monitoring tools to predict anxiety using wearable devices.