Smartwatch-Based Human Stress Diagnosis Utilizing Physiological Signals and LSTM-Driven Machine Intelligence
Authors:
- 3 June 2024
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 01-08
Abstract
Stress responses, primarily the ‘fight-or-flight’ reactions, are fundamental to human survival but can become harmful when they persist for extended periods. Prolonged stress exposure has been linked to various health issues and accelerated aging. But this can be addressed or controlled if a person becomes aware of their stress condition early. Traditional methods of identifying human stress relied heavily on self-reported questionnaires, but with the advancements in machine intelligence utilizing Machine Learning (ML) and Deep Learning (DL) techniques, the use of smart wearables like smartwatches for diagnosing stress has gained prominence. These devices can collect real-time physiological data, e.g., heart rate, skin conductance, and movement patterns, which are crucial for detecting stress. In this study, we utilized the ‘Stress-Predict Dataset’ built from the Empatica E4 wearable smartwatch device. After processing raw signals collected from 35 participants and aligning them to a uniform frequency of 32 Hz, we developed a Long Short-Term Memory (LSTM) model for stress classification. Training and validation sets achieved accuracies of 93.67 % and 91.13 %, respectively. The test set achieved an accuracy of 91.78 %, indicating a strong ability to identify stress events. This includes correctly classifying 30.31 % of True Positive (TP) stress occurrences with a low rate of False Negatives (FN) (missed stress events) at 4.56%. There were also 61.47% of True Negative (TN) classifications, demonstrating the model's ability to correctly identify non-stressful states with a low rate of False Positives (FN) (incorrectly classified non-stress events) at 3.66%. This research contributes to the ongoing exploration of stress diagnosis using wearable smartwatch device data, underlining the potential for further advancements in physiological stress monitoring.Keywords
This publication has 27 references indexed in Scilit:
- Ensemble machine learning model trained on a new synthesized dataset generalizes well for stress prediction using wearable devicesJournal of Biomedical Informatics, 2023
- An Individual-Oriented Algorithm for Stress Detection in Wearable Sensor MeasurementsIEEE Sensors Journal, 2023
- Comparing stress prediction models using smartwatch physiological signals and participant self-reportsComputer Methods and Programs in Biomedicine, 2021
- A Review on Mental Stress Detection Using Wearable Sensors and Machine Learning TechniquesIEEE Access, 2021
- Array programming with NumPyNature, 2020
- Comparative Study on the Effect of Order and Cut off Frequency of Butterworth Low Pass Filter for Removal of Noise in ECG Signal2020
- Transductive LSTM for time-series prediction: An application to weather forecastingNeural Networks, 2020
- Hybrid speech recognition with Deep Bidirectional LSTM2013
- Application of LSTM Neural Networks in Language Modelling2013
- Perceived Stress Scale1983