A Two-Stage Anomaly Filter Model for Continuous Stress Monitoring-Based on Ppg

Stress plays a significant role in human well-being. Chronic stress can lead to many life-threatening diseases like hypertension, diabetes, and asthma. Therefore, efficient monitoring and mitigating stress are of prime interest among researchers in this domain. The advancement of wearable and sensor technologies has made it possible to develop continuous stress monitoring in real-time. This study emphasizes developing a continuous stress estimator dedicated to wearable devices or smart watches. The study leverages the time domain features derived from inter-beat-interval signals from a photo photoplethysmography (PPG) sensor. A two-stage data filter is employed that uses an anomaly filter followed by a median filter. The features derived from the filtered data are utilized to get a stress estimation model. The proposed stress estimator has achieved a root mean squared error of 0.2 which is found lowest compared to the other baseline models. The model is evaluated in the edge device in the context of memory and inference time capabilities.