A Wearable Electronic Band for Stress Understanding Using Machine Learning

Stress is a normal aspect of life, but when it persists for too long, it can significantly impact mental health. Prolonged stress is a serious issue that can deeply impact daily life and mental well-being. Recognizing the signs of chronic stress and taking proactive steps to manage it can help prevent the development of mental disorders, fostering a healthier, more balanced life. There is a growing need for wearable devices to monitor daily life, especially stress management and overall mental health. Integrating Deep Learning (DL) with wearable devices can significantly enhance the effectiveness of continuous, real-time feedback on physiological and behavioral metrics, empowering individuals to take proactive steps toward better mental health. This work enables the acquisition of Heart Rate (HR) from Photoplethysmography (PPG) and Electrodermal Activity (EDA) from Galvanic Skin Resistance (GSR) sensor to monitor stress using a designed wearable watch. Instantaneous Frequency (IF) features from physiological signals namely Interbeat Interval (IBI) derived from HR and Skin Conductance Level (SCL) from Electrodermal Activity (EDA) were extracted from Ensemble Empirical Mode Decomposition (EEMD) based Hilbert Transform (HT). Three datasets were generated by labeling participant self-reports and analyzing the power of stress-associated frequencies derived from IF characteristics of physiological signals. Bidirectional Long Short-Term Memory (Bi-LSTM) based on Deep Learning (DL) is deployed to perform binary classification achieving an accuracy and F1-score of 99.38% and 98.88% respectively by dataset collected in this work. This work also attempted to generate labels from the extracted features, which can be implemented in real-time scenarios where collecting reports is impossible.