An Automatic Stress Analysis and Emotion Detection Framework for Humans Based on Several Brain Signals

Abstract

In common parlance, stress is the feeling people get when they have a lot on their plates and are having trouble keeping up with everything. Although some degree of stress might be useful in the short term, chronic stress has serious negative effects on health, including increased risk of cardiovascular problems such heart disease, high blood pressure, and stroke. Depression, anxiety, and personality disorders are also possible outcomes. Hence, stress detection is helpful for managing stress-related health problems. Perceptual, behavioural, and physiological reactions all provide useful indicators of stress levels. Based on feature extraction and classification methods, numerous researchers have presented a wide variety of methods. Some of these methods are difficult to implement and produce unreliable outcomes when applied to the study of human stress.
Patient denial, insensitivity, subjective biases, and inaccuracy are only some of the issues that arise from relying solely on doctor-patient interaction and scale analysis when diagnosing Stress. To better predict clinical results, a computerised, objective system is needed for Stress diagnosis and treatment. In an effort to better detect Stress, this research modifies EEG data and use machine learning algorithms. Ten participants' EEGs were recorded using a Narosky system while they were exposed to various emotional face cues. Psychologists relied on the EEG signal as a diagnostic tool for Stress. Both machine learning and deep learning were used to analyse the features. Using PCA, ICA, and EMD for BCI applications yields significant results. With SVM, a programmer can reap many benefits: PCA has great generalisation properties, and it can detect tension and pressure from EEG signals. The effect of overtraining is particularly vulnerable to the curse-of-dimensionality when the signals are negative. By analysing EEG waves, we were able to detect Stress with these benefits. The experimental study provides a somewhat comprehensive summary of the various methods, all of which rely on frequency domain analysis of 14 EEG data.