Analysing Effectiveness of Different Physiological Biomarkers in Detecting Stress

Stress is a common part of our everyday life and helps us to handle certain situations in a better way. However, exposure to stress over a long period of time can cause mental and physical health complications. Many health complications associated with stress can be prevented with its early detection. Stress in humans is accompanied by certain physiological changes and monitoring those changes can help in its detection. In this, we study the impact of different physiological changes on stress detection. Different machine learning and deep learning techniques viz. Random Forest, AdaBoost, Decision Trees, Gaussian Naive Bayes and Artificial Neural Networks were used for stress detection using WESAD dataset, which is a dataset for stress and affect detection. Artificial Neural Network outperformed the machine learning techniques and achieved an accuracy of 93.6%. To analyse the effectiveness of each physiological biomarker in detecting stress, the artificial neural network was trained separately on every possible combination of biomarker set. The series of experiments showed that only three physiological biomarkers viz. three axes acceleration, electrodermal activity and body temperature, are sufficient to detect stress effectively.