Objective Stress Detection in Peru via a Multimodal Wearable System of ANS Biomarkers

Mental health disorders represent a growing chal- lenge in Latin American university populations, where stress assessment relies primarily on self-report questionnaires, limiting continuous monitoring. This study presents the design and vali- dation of a non-invasive multimodal wearable system for stress detection using autonomic nervous system (ANS) biomarkers, acquired from 7 students under a standardized acute stress induction protocol. The high sampling resolution yielded 165,744 temporal instances, labeled via the Visual Analog Scale (VAS) into three categories (Mild, Moderate, and Intense) and classified using supervised learning over sliding temporal windows. The Random Forest model achieved the best performance with 81.37% accuracy, AUC of 0.9460, and Recall of 0.8108, with a 95% confidence interval [73.81% – 88.93%], confirming the system’s viability for objective stress detection in Latin American academic settings.