Pulse2stress: HRV-Based Stacked LSTM-AE Framework for PPG Stress Detection

This study proposes a Stacked LSTM-AE-MLP model for detecting stress intervals from PPG (Photoplethysmography) signals using HRV (Heart Rate Variability) time-series information extracted from 3-minute windows. Prior PPG-based stress studies have predominantly relied on independent classification of short window segments, limiting their ability to capture the gradual, continuous nature of stress dynamics. To address this limitation, the proposed model incorporates HRV, enabling the representation of midterm temporal patterns underlying stress accumulation and alleviation. Furthermore, the LSTM-AE architecture learns and reconstructs latent HRV sequence patterns, facilitating robust representation learning. Subject-wise evaluation on the WESAD dataset demonstrates 93.75 % accuracy and stable reconstruction performance, indicating the model's generalization across individuals with varying physiological characteristics. These findings highlight the potential of the proposed approach as a foundation for practical biosignalbased stress detection and monitoring systems.