Scholarlink

<p>Stress has become a major contributor to various diseases and health issues, prolonged exposure can lead to a shorter lifespan and increased mortality. Hence detecting the stress in its early stages will minimize the risk of severe or life-threatening ailments. Therefore, it is meaningful to do experiments on the physiological signals of humans through wearables that can detect or indicate the stress state. This paper investigates the physiological signals from photoplethysmography (PPG) and Galvanic Skin Resistance sensors of K-EmoCon dataset for machine learning-based stress detection, focusing on Heart Rate Variability (HRV) and Electrodermal Activity (EDA) as the recorded signals. The novelty of the work is the signal processing method used to extract the signature of the stressed and non-stressed pattern of HRV and EDA signals. Time–frequency content of these signals was extracted through Ensemble Empirical Mode Decomposition technique to get Intrinsic Mode Function (IMF). When combined with Hilbert Huang Transform for feature engineering, we derived the Instantaneous Frequency (IF), Instantaneous Amplitude (IA) and Marginal Spectrum (MS) from IMF. Considering the IF, IA and MS as extracted features from the physiological signals we performed binary classification using Support Vector Machine (SVM) classifier, achieving outstanding results: 98.44% accuracy, 98.36% precision, 99.84% recall, 99.09% F1-score, and 98.66% Area Under ROC (Region of Convergence Curve (AUC). Signal processing for feature extraction and optimized SVM model construction resulted in significant improvement of 11.31% with the original work of Zitouni et al. (IEEE J Biomed Health Inform 27(2):912–923, 2022). The proposed SVM model is simple and suitable for real-time monitoring solutions with substantial potential for medical technology and healthcare applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning-Based Human Stress Detection Model Employing Physiological Sensory Data

  • 期刊论文

  • 发表日期：

    2025年1月21日

查看全文
  • Subathra Panneer Selvam,
  • Malarvizhi Subramani,
  • Ferents Koni Jiavana,
  • Arul Saravanan Ramachandran

摘要

Stress has become a major contributor to various diseases and health issues, prolonged exposure can lead to a shorter lifespan and increased mortality. Hence detecting the stress in its early stages will minimize the risk of severe or life-threatening ailments. Therefore, it is meaningful to do experiments on the physiological signals of humans through wearables that can detect or indicate the stress state. This paper investigates the physiological signals from photoplethysmography (PPG) and Galvanic Skin Resistance sensors of K-EmoCon dataset for machine learning-based stress detection, focusing on Heart Rate Variability (HRV) and Electrodermal Activity (EDA) as the recorded signals. The novelty of the work is the signal processing method used to extract the signature of the stressed and non-stressed pattern of HRV and EDA signals. Time–frequency content of these signals was extracted through Ensemble Empirical Mode Decomposition technique to get Intrinsic Mode Function (IMF). When combined with Hilbert Huang Transform for feature engineering, we derived the Instantaneous Frequency (IF), Instantaneous Amplitude (IA) and Marginal Spectrum (MS) from IMF. Considering the IF, IA and MS as extracted features from the physiological signals we performed binary classification using Support Vector Machine (SVM) classifier, achieving outstanding results: 98.44% accuracy, 98.36% precision, 99.84% recall, 99.09% F1-score, and 98.66% Area Under ROC (Region of Convergence Curve (AUC). Signal processing for feature extraction and optimized SVM model construction resulted in significant improvement of 11.31% with the original work of Zitouni et al. (IEEE J Biomed Health Inform 27(2):912–923, 2022). The proposed SVM model is simple and suitable for real-time monitoring solutions with substantial potential for medical technology and healthcare applications.