Stress Detection using Wearable Physiological Sensors and Machine Learning Algorithm

The primary objective of this methodology is to develop and test the performance of a wearable psychological sensor using EEG, EDA and ECG to observe the stress level of humans and analyze if the observed variation in psychological signals is related to the objective biomarker of stress. An integrated wearable system with physiological sensors is designed, developed, tested and evaluated in this paper for monitoring stress using biological markers in the working environment. Stress detection is performed using the Muse S (Gen 2) EEG headset, Savvy ECG sensor, and Shimmer3 GSR sensors. For each subject under test, 32 features are extracted from the multi-modal signals, of which, five features are retained for EEG, four signals for ECG and 1 for EDA, summing up to 10 features. The window size for the collection of each feature is one minute. Hence, for the 20 participants and ten features, a data of 200 minutes is used as training data in a controlled environment. Of these, 138 minutes are labelled as stressful tasks and 62 minutes are labeled as no-stress. The proposed stress detection model effectively assesses stress under a controlled environment with an accuracy of 97% and in an everyday environment with an accuracy of 93%. This work aims to design and develop a remote control system that can be used in medical devices to prevent the aftereffects of stress.