Early Stress Detection and Analysis using EEG signals in Machine Learning Framework

Abstract

Abstract Stress, a psychological phenomenon that represents the body’s natural defense against predators and danger, has emerged as the biggest social problem of the 21st century especially during the Covid-19 pandemic. Various techniques or methods such as PET, ECG, EMG, MRI exist to detect and quantify stress. Physiological features produced throughout the brain’s electrical activity are documented by a medical technique known as an electroencephalogram (EEG). In this context, this paper posits a comparative analysis of the above-described methods of stress detection and accentuates on stress detection methodology using EEG signals, as EEG is a perfect non-invasive tool, widely used in clinical and research domains. The fractal dimension (FD) method, which is an indicator of curve irregularities, has been used in the detection of stress for feature extraction, applying three FD algorithms viz. Higuchi, Katz and Permutation Entropy. For classification, this study aims to apply and compare a number of classic machine learning algorithms based on accuracy, precision and sensitivity. This paper also presents a novel architecture, based on EEG analysis in MATLAB, fractal dimension used for feature extraction along with Machine Learning processes for classification i.e., Random Forest and Artificial Neural Network which is useful for early-stage stress detection, analyzing different stress levels viz. mild, moderate and high accuracy and providing ways for people to cope with stress in order to enhance their performance.