Stress Detection with Machine Learning and Deep Learning using Multimodal Physiological Data

Stress is a common part of everyday life that most people have to deal with on various occasions. However, having long-term stress, or a high degree of stress, will hinder our safety and disrupt our normal lives. Detecting mental stress earlier can prevent many health problems associated with stress. When a person gets stressed, there are notable shifts in various bio-signals like thermal, electrical, impedance, acoustic, optical, etc., by using such bio-signals stress levels can be identified. This paper proposes different machine learning and deep learning techniques for stress detection on individuals using multimodal dataset recorded from wearable physiological and motion sensors, which can prevent a person from various stress-related health problems. Data of sensor modalities like three-axis acceleration (ACC), electrocardiogram (ECG), blood volume pulse (BVP), body temperature (TEMP), respiration (RESP), electromyogram (EMG) and electrodermal activity (EDA) are for three physiological conditions - amusement, neutral and stress states, are taken from WESAD dataset. The accuracies for three-class (amusement vs. baseline vs. stress) and binary (stress vs. non-stress) classifications were evaluated and compared by using machine learning techniques like K-Nearest Neighbour, Linear Discriminant Analysis, Random Forest, Decision Tree, AdaBoost and Kernel Support Vector Machine. Besides, simple feed forward deep learning artificial neural network is introduced for these three-class and binary classifications. During the study, by using machine learning techniques, accuracies of up to 81.65% and 93.20% are achieved for three-class and binary classification problems respectively, and by using deep learning, the achieved accuracy is up to 84.32% and 95.21% respectively.