Stress Detection from Multimodal Wearable Sensor Data
Fitri Indra Indikawati, Sri Winiarti
Abstract
Abstract Stress can be recognised by observing changes in physiological responses on the human body. Wearable sensors for stress detection are becoming more prominent in recent years due to their functionality and non-intrusive nature. By utilising data from wearable sensors, we have developed a personalized stress detection system. Our system performs classification on stress level using multimodal data from wrist-worn device Empatica E4 wearable sensor. We implemented three different classification algorithms: Logistic Regression, Decision Tree, and Random Forest and used four-class classification conditions: baseline, stress, amusement, and meditation. By evaluating the performance of the system, we demonstrate that our system can perform the best and consistent personalized stress detection using Random Forest classifier with the accuracy of 88%-99% on 15 subjects.