Predictive Analysis of Mental Stress using Machine Learning Techniques

Abstract

Many stress-related health problems can be avoided if mental stress is identified early. It is critical to have reliable techniques for detecting human stress quickly and accurately. This study aims to measure physiological indicators and recognize mental fatigue using a wrist-worn sensor gadget. A group of healthy individuals were exposed to three distinct stresses. The electrocardiogram, breathing, skin conductivity, and EMG of the trapezius muscles were all monitored throughout the operation. These data were used to calculate 19 physiological parameters. A subset of nine characteristics was chosen for further analysis once the feature values had been normalized, and correlations between these features were examined. Through the use of principal component analysis, these nine characteristics were reduced to seven principal components. It was discovered that by using these primary components in conjunction with a number of classifiers, a classification accuracy of approximately 90% could be maintained between stressful and non-stressful conditions. This indicates that a potentially useful feature subset may have been discovered, which can be used in the future to construct a personalized stress monitor. When using these strategies, obtaining the best characteristics is never going to be an easy task. As it focuses on the binary classification problem, this work will employ the Bernoulli Naive Bayes method, one of the best classification algorithms. An attempt is made to identify the optimal set of features that maximizes classification accuracy.