Unveiling Stress Responses through Heart Rate Variability Analysis: A Machine Learning Approach

Stress has been called the "hidden epidemic" since it affects every aspect of human life. A prerequisite for effective stress management is early detection. Advances in wearable sensor technologies have led to various approaches for detecting stress using physiological signals. In this study, we utilized a publicly available data set from PhysioNet which contains data collected from 10 participants during three exams (midterm 1, midterm 2, and final) using the Empatica E4 device. For each subject, we employed HRVanalysis, a Python package, to convert interbeat interval data recorded for each condition into time domain and frequency domain heart rate variability features. We developed a ternary Multi-Layer Perceptron implementation of the Artificial Neural Network architecture that employed Grid Search for hyper-parameter optimization and five-fold stratification for cross-validation to classify participants' data into low, medium, and high stress. Our model achieved 83% accuracy, 89% precision, 83% recall, and 82% F1-score, indicating its ability to detect stress.