HRV and GSR as Viable Physiological Markers for Mental Health Recognition
Shikha Shikha, Aryan Aryan, Lovish Arya, Divyashikha Sethia
Abstract
Mental stress has become a standard part of day-to-day life. However, experiencing long-term and high-level stress affects the daily life and wellness of the person. Consequently, an individual's performance and management ability degrade significantly in critical situations. Electrocardiogram (ECG), Galvanic Skin Response (GSR), Electromyogram (EMG), Skin Temperature (ST), and Respiration are essential physiological biomarkers to quantify stress effectively. This paper aims to classify the stress level with improved performance based on GSR and ECG-derived Heart Rate Variability (HRV) features using machine and deep learning algorithms. It uses the Stress Recognition in Automobile Drivers (SRAD) dataset, which contains a collection of multiparameter recordings from 17 healthy participants who drive on a prescribed route under various pressure conditions. The work uses Pearson's Correlation, RFECV, and LightGBM feature selection methods with different classifiers to reduce redundancy between features and enhance performance. The accuracy and F1-score for stress level classifications are computed and compared using machine and deep learning algorithms. For binary classification (stress vs. non-stress), Random Forest achieves the best classification accuracy of 93.96 % which is higher than previous works. It also provides an accuracy of 81.41 % for three-class (baseline vs. medium stress vs. high stress) stress level classification.