Stress is a natural physiological and psychological response to external demands or perceived threats. However, when stress becomes chronic or excessive, it significantly heightens the risk of mental and physical health disorders, including anxiety, depression, and sleep disturbances. A dependable physiological indicator for stress evaluation is heart rate variability (HRV), which shows the fluctuation in the time intervals between subsequent heartbeats (RR intervals). Heart rate variability (HRV) records the dynamic changes in cardiac activity, providing greater insights into autonomic nervous system function, as opposed to heart rate, which offers an average beats-per-minute assessment. Here, we present a CNN-based deep learning model that investigates the possibility of HRV features as biomarkers for multi-class stress detection. A lack of stress, stress caused by interruptions, and stress caused by time pressure are the three types of stress that the model is intended to classify. In order to improve the accuracy of detection, HRV is used in both the time-domain and frequency-domain. The suggested model is tested on the open-source SWELL-KW dataset and shows remarkable results: a 99.9% accuracy rate, a 1.0 precision score, a 1.0 recall score, a 1.0 F1-score, and a 0.99 Matthews Correlation Coefficient (MCC). In addition, an analysis of variance (ANOVA) feature extraction method validates the efficacy of critical HRV features. The results show that deep learning methods based on HRV are capable of reliable multi-class stress identification.
Charitha et al. (Tue,) studied this question.