AI for health: Using AI to identify stress from wearable devices data

Stress is an escalated psycho-physiological state of the human body emerging in response to a challenging event or a demanding condition and since it is one of the major issues in modern society, there is a growing interest in developing methods that make automatic detection possible. To this end, the adoption of wearable technology coupled with the implementation of machine learning (ML) techniques are emerging as an interesting approach to develop non-invasive stress detection systems. The present work investigates the coupled use of Machine Learning and biosignals collected from wearables in a controlled environment to study the feasibility of non-invasive stress detection systems. This thesis uses a dataset that was obtained by acquiring three different biosignals from subjects during a stress test by using the BiosignalsPlux platform: Electrocardiogram (ECG), Electrodermal Activity (EDA) and Respiration Signal. The collected data was then pre-processed and prepared by applying a features extraction and features selection algorithms in order to work with tabular data. The selected features are prepared for use in multiple AI algorithms, including Random Forest, XGBoost, Neural Network and Support Vector Machine. The results of the thesis showed that machine learning algorithms were successful in detecting stress from biosignals with high accuracy level and it also shows that the use of multiple biosignals is more effective compared to single-signal-based system. XGBoost was the best performing algorithm achieving an accuracy of 84% for the binary classification (stress vs. no stress) and 70% for three class classification (no stress, medium stress and high stress). The findings of this work may have important implications for the development of non-invasive wearable devices for early stress detection and management.