Estimating Psychophysiological Changes Using Heart Rate Variability Analysis

Abstract

The autonomic nervous system (ANS) plays a crucial role in maintaining physiological balance, responding dynamically to external modulators categorized as stressors and relaxers. Heart Rate Variability (HRV) serves as a key biomarker for assessing these influences, enabling objective analysis of stress and relaxation states. This research presents a multi-domain approach to HRV analysis, integrating standard time-domain and frequency-domain features with advanced techniques such as Fuzzy Recurrence Plot (FRP) and Empirical Mode Decomposition-based FRP (IMF_FRP_GLCM). First, the study investigates the impact of city driving stressors and slow-paced breathing relaxers on ANS activity, revealing that RMSSD consistently exhibits opposite trends under these conditions, making it a robust indicator of stress and relaxation states. Furthermore, to deepen the understanding of the autonomic responses, the study converts HRV time series into image-based representations—particularly through fuzzy recurrence plots. This transformation adds a powerful visual dimension that preserves temporal and nonlinear dynamics, making subtle physiological changes associated with relaxation more apparent. By visualizing these patterns, clinicians and individuals can more intuitively observe and track autonomic responses to slow-paced breathing, thereby enhancing interpretation, communication, and adherence in relaxation-focused interventions. Building upon this, the study explores time-series-to-image conversion techniques to enhance HRV classification, leveraging FRP for improved feature extraction and machine learning performance. HRV time series collected from 60 participants during spontaneous and slow paced breathing were analyzed across different segment lengths. Results demonstrate that standard HRV features provide optimal classification performance for 5-minute segments, while IMF_FRP features maintain high accuracy even for ultra-short segments, aligning with real-time monitoring requirements in wearable health devices. Feature selection methods such as Fisher Discriminant Ratio (FDR) and greedy search improved classification efficiency, with Support Vector Machine (SVM) achieving an accuracy of 96.6% and specificity of 100% for 5-minute segments. The findings underscore the significance of FRP-based analysis in detecting physiological states and provide a foundation for integrating HRV-based stress and relaxation detection into smart wearable technology. By bridging traditional HRV analysis with novel machine learning techniques, this research advances the objective and data-driven methods for stress monitoring and personalized health interventions.