Distinguishing Stressor, Stress, and State-Anxiety: Semantic and Physiological Insights with Machine Learning Approaches

Stress and state anxiety are natural defense mechanisms of the human body, aiding in adaptation to various scenarios and playing a crucial role in human survival. According to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), untreated stress and state anxiety can evolve into pathological conditions such as post-traumatic stress disorder, generalized anxiety disorder, and depression. Diagnosing these conditions typically involves professional interviews, which can be challenging due to the overlap of symptoms with other conditions, the periodic nature of these assessments, and the lack of continuous mental health monitoring. Thus, developing objective metrics for early identification and constant monitoring of stress and state anxiety is essential to prevent health deterioration and improve treatment outcomes. This work aims to establish a robust methodology for analyzing and classifying stressors, stress, and state-anxiety using machine-learning models. It identifies each condition’s semantic differences and physiological impacts through signals such as heart rate, galvanic skin response, and blood volume pressure. It also introduces two convolutional network architectures: the Single-Input model, which evaluates the individual contribution of each signal, and the Multi-Input model, designed for inputs from multiple sensors with different sampling frequencies. Additionally, it proposes a new validation setup called Repeated Leave-One-Subject-Out Cross-Validation (RepeatedLOSOCV) to yield more precise results by considering intra- and inter-individual biological variations with small datasets.