Comparison of Machine Learning Approaches for Physiological States Classification Using Heart Rate and Pulse Rate Variability Indices

Aug 6, 2026

Upload by Logic

This work evaluates the feasibility of classifying physiological states related to postural and mental stress by using heart rate variability (HRV) and pulse rate variability (PRV) indices computed in time, frequency, and information domains. Four machine learning algorithms—LDA, support vector machines, neural networks, and k-nearest neighbors—are compared with and without mRMR feature selection. Experiments on 76 healthy young subjects across supine rest, orthostatic, and mental stress show higher HRV accuracy than PRV and best discrimination for orthostatic stress.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages4

Reading Time10 min

AccessFree to download

Introduction

Materials and Methods

Subjects and experimental protocol

Comparison of Machine Learning Approaches for Physiological States Classification Using Heart Rate and Pulse Rate Variability Indices Marta Iovino 1 , Michal Javorka2 , Luca Faes 1 , and Riccardo Pernice 1 1 Department of Engineering, University of Palermo, Building 9, Viale delle Scienze, Palermo, Italy 2 Department of Physiology, Jessenius Faculty of Medicine, Comenius University, Martin, Slovakia Abstract—In this work, we investigate the feasibility of classifying physiological states including conditions of postural and mental stress using heart rate variability (HRV) and pulse rate variability (PRV) time-, frequency-and information-domain indices. The performance of four different machine learning algorithms, i.e. Linear Discriminant Analysis (LDA), Support Vector Machines, Neural Networks (NN) and k-Nearest Neighbors, were compared, with and without prior applying the minimum Redundancy Maximum Relevance (mRMR) algorithm for feature selection. Analyses were conducted on 76 young healthy subjects under three different conditions (supine rest, orthostatic and mental stress). Results evidence higher accuracy for HRV indices if compared to PRV and better classification performance of orthostatic stress. The highest accuracy has been achieved by the NN algorithm on HRV time series, with a value of 90.9% after feature selection, while LDA is the best algorithm for PRV features (81.8%). Keywords—Heart Rate Variability (HRV), stress classification, feature selection, Machine Learning (ML). I. INTRODUCTION In recent years, the application of machine learning (ML) techniques to medical data analysis has significantly increased in the research field and healthcare industry. ML has been especially exploited to help to classify various autonomic nervous system states related to different types of stress [1], their excess being considered one of the most important pathogenic factors of the modern life [2] . Usually, noninvasive stress assessment is carried out by studying Heart Rate Variability (HRV), i.e. the beat-to-beat variation of heart rate (HR), and its indices represent the most reliable markers of mental and physical stress [3] . HRV reflects the complexity of the cardiovascular regulation and, given the inhibition of parasympathetic accompanied by the activation of the sympathetic branch of the autonomous nervous system during stress conditions, can reflect the body’s ability to respond to various (e.g. environmental and psychological) stimuli [3] . Short-term HRV is usually assessed by measuring the time interval between consecutive heartbeats (i.e. R-R intervals) from 5-minute recordings of electrocardiographic (ECG) signals (i.e. short-term HRV) and computing time-, frequency- and information-domain measures [3],[4] . Recently, there has been an increased interest in studying whether and to what extent HRV can also be studied from photoplethysmographic (PPG) or blood pressure recordings, providing the so-called Pulse Rate Variability (PRV) . PPG is an optical technique able to detect changes in microvascular blood volume in tissues and is widely used in wearable devices, being simple, low-cost, safe and minimally invasive [5] . Even if the two techniques (PPG and ECG) are often considered interchangeable to measure HRV, several reasons suggest that the beat-to-beat variability recorded with the PPG is somewhat different from HRV [5],[4], [6] . PPG and blood pressure recordings can be affected by physiological factors related to the transmission of the pulse wave through the vascular system and measurement errors due to motion-induced signal corruption, lowering the peak detection accuracy and thus reducing the agreement between PRV and HRV. Applying machine learning techniques to HRV and PRV features to classify the stress level thus represents nowadays an important challenge for researchers. In particular, various studies have focused on classifying physiological states using HRV [7], [8] or PRV [1] t

FAQs

Which physiological states are classified in the study?

The study classifies postural and mental stress states using three experimental conditions: supine rest, orthostatic stress, and mental stress.

How are HRV and PRV features computed for machine learning?

Features are derived from indices in time, frequency, and information domains from HRV (beat-to-beat R-R intervals) and PRV (pulse-derived recordings such as PPG/blood pressure).

Which machine learning approach achieves the best accuracy for each feature type?

The highest accuracy for HRV time series is obtained with neural networks (90.9%) after feature selection, while LDA provides the best results for PRV features (81.8%).

Aug 6, 2026

Upload by Logic