Near Real-time Stress Prediction for Patients with Disturbed Allostatic Load

Abstract

Stress is one relevant cause of diseases nowadays, and prolonged exposure to stress can cause a disturbance in the allostatic load. Alternatives have been sought to deal with this situation and verify the impact of this allostatic load disorder. Wearable sensors are an option for automatically identifying acute stress since they can measure signs such as electrocardiogram, heart rate, electroencephalogram, electromyogram, or galvanic skin response. All these signals have intrinsic characteristics in a normal state and change if associated with stress occurrence. The literature presents Machine Learning Approaches and Deep Learning Models as alternatives to pattern detection in physiological signals. Nevertheless, we identify a gap regarding the allostatic load impact identification and the real-time classification when using these models. this article aims to acquire data in stress induction experiments in clinical and non-clinical patients, train a machine learning model, and, in sequence, carry out a new experiment to evaluate the classification in near real-time. The classification experiment presented results with accuracy above 92.72%. When it comes to real-time classification experiments we obtained an accuracy of 78.93%. Evaluating participants in experiments divided into clinical and non-clinical groups, a decrease of 5% in precision was identified. Based on the results obtained, we verified that the allostatic load can present challenges for real-time stress classification.