Machine Learning Model for Stress and Anxiety Recognition Using Heart Activity Signals.

Numerous recent studies have focused on using physiological signals from body-worn sensors to detect negative emotional states like stress and worry. Many authors have achieved good accuracy findings using a variety of signals, such as cardiac, skin conductance, and skin temperature. Machine learning classifiers are typically used to features acquired from sensor data. However, it is rarely discussed how reliable these models are for field implementation. In this work, we assess the generalizability of models built from cardiac signals with an emphasis on stress and anxiety detection using publicly available data from two extensive experimental studies. Our findings also highlight how crucial it is to incorporate a variety of emotional states within the training set in order to reduce incorrect classification from hidden areas.