Machine Learning and Deep Learning approaches for stress detection using Empatica E4 bracelet

In response to challenging circumstances, the human body can experience marked levels of anxiety and distress. In order to prevent stress-related complications, timely identification of stress symptoms is crucial, necessitating the need for continuous stress monitoring. Wearable devices offer a means of real-time and ongoing data collection, facilitating personalized stress monitoring. This study aimed to detect stress by analyzing physiological signals collected through the Empatica E4 bracelet. Machine Learning algorithms (Random Forest, SVM, Logistic Regression) and Deep Learning pre-trained CNNs (GoogLeNet, SqueezNet) were employed to differentiate between stressful and non-stressful situations. Data from 29 subjects, including photoplethysmographic (PPG) and electrodermal activity signals (EDA), were used to extract 27 features with and without overlapping. These features were then utilized in three Machine Learning algorithms for binary classification using Python, after applying the Chi-square test and Pearson’s correlation coefficient via WEKA for feature importance ranking. Additionally, SHapley eXplainable AI was applied to the top-performing model, Random Forest, in the overlapping case, shedding light on the most impactful features and comparing them with feature selection methods. Notably, HRV (Heart Rate Variability) features emerged as significant in stress detection. Furthermore, in the non-overlapping case, continuous wavelet transform was applied to PPG signals to generate scalograms, which were subsequently fed into two different pre-trained CNNs. The study’s results showcased the overlapping had a positive impact on all models. Moreover, the Random Forest model is the highest-performing, achieving an accuracy of 76.4% without overlapping and an impressive 99.5% with overlapping segments. Additionally, Deep Learning models exhibited potential in stress classification, particularly when considering the use of PPG signals only.