International Journal of Scientific Research and Engineering Development

Wearable photoplethysmography (PPG) offers scalable monitoring of short-timescale physiological stress, but motion artifact and the non-equivalence of pulse-rate variability (PRV) and electrocardiographic heart-rate variability (HRV) limit reliability. This study evaluates a quality-aware PPG machine-learning framework using the WESAD benchmark. Wrist blood-volume pulse (64 Hz), wrist accelerometry (32 Hz), and chest ECG (700 Hz) from 15 participants were segmented into 865 60-s windows with a 30-s step for baseline-versus-stress classification. Subjectindependent leave-one-subject-out (LOSO) evaluation compared raw PRV, conventionally filtered PRV, PPG+ACC, quality-aware PRV, and a full PPG+ACC+quality model. Raw PRV produced the highest mean Macro-F1 (0.8379), whereas PPG+ACC achieved the highest AUROC (0.9560) and AUPRC (0.9330). The full model achieved AUROC 0.9524, AUPRC 0.9295, and the lowest Brier score (0.1164). Selective prediction reduced error from 13.33% at 79.77% coverage to 7.48% at 61.85% coverage, while accepted-window Macro-F1 increased from 0.8346 to 0.8885. ECG-referenced analysis showed strong correlation for heart rate (r = 0.841, p = 0.0107) but weak correlation and poor agreement for RMSSD (r = 0.209; bias = 0.1669 s), demonstrating that PPG-derived PRV should not be treated as interchangeable with ECG-derived HRV. The results support uncertainty-aware wearable stress-state monitoring while identifying physiological validity and between-subject heterogeneity as key deployment constraints.

Egberibo Paul Waredibamo, Oluwaseun Omotoso,"Quality-Aware Wrist Photoplethysmography for Subject Independent Stress-State Monitoring: ECG-Referenced Validation and Selective Prediction" International Journal of Scientific Research and Engineering Development, V8(6): Page(3281-3289) Nov-Dec 2025. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.