Scholarlink

<p>In this challenging era, psychological stress has become a pervasive issue affecting human well-being. Timely identification of stress, especially through physiological signals such as Electrocardiograms (ECG), is essential for early intervention and effective stress management, as these signals offer objective, real-time insights into autonomic nervous system responses that are closely linked to stress levels. While existing research largely focuses on binary classification (e.g., stressed vs. non-stressed) or multi-level classification restricted to specific domains such as workplace environments or academic settings, there remains a gap in developing a generalized model applicable across diverse real-world conditions. This study introduces a novel Multi-Level Attentive LSTM Network (MALN) for classifying real-world ECG signals into three stress levels: normal, low, and mid. The approach leverages the robustness of deep learning in analyzing complex biomedical time-series data. ECG signals from drivers (a high-stress occupational domain) and graduate students (an academic stress domain) are transformed into time-dependent HRV and spectral sequences to preserve meaningful temporal and frequency-based characteristics. Two attention-guided Bi-directional LSTM modules are designed to extract relevant patterns from these sequences. Their outputs are concatenated and passed through a fully connected layer to distinguish the features into three stress levels. Experimental evaluations conducted on a public drivers’ ECG dataset and a custom-built student ECG dataset show that the proposed model effectively generalizes across distinct domains and outperforms state-of-the-art methods in multi-level stress classification. This work is available at <a href="https://github.com/Ramyashri14-0593/Stress-Level-Classification">https://github.com/Ramyashri14-0593/Stress-Level-Classification</a>.</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

Discriminating psychological stress levels: multi-level attentive LSTM approach

  • 期刊论文

  • 发表日期：

    2025年9月13日

查看全文
  • Ramyashri B. Ramteke,
  • Gaurav O. Gajbhiye,
  • Vijaya R. Thool

摘要

In this challenging era, psychological stress has become a pervasive issue affecting human well-being. Timely identification of stress, especially through physiological signals such as Electrocardiograms (ECG), is essential for early intervention and effective stress management, as these signals offer objective, real-time insights into autonomic nervous system responses that are closely linked to stress levels. While existing research largely focuses on binary classification (e.g., stressed vs. non-stressed) or multi-level classification restricted to specific domains such as workplace environments or academic settings, there remains a gap in developing a generalized model applicable across diverse real-world conditions. This study introduces a novel Multi-Level Attentive LSTM Network (MALN) for classifying real-world ECG signals into three stress levels: normal, low, and mid. The approach leverages the robustness of deep learning in analyzing complex biomedical time-series data. ECG signals from drivers (a high-stress occupational domain) and graduate students (an academic stress domain) are transformed into time-dependent HRV and spectral sequences to preserve meaningful temporal and frequency-based characteristics. Two attention-guided Bi-directional LSTM modules are designed to extract relevant patterns from these sequences. Their outputs are concatenated and passed through a fully connected layer to distinguish the features into three stress levels. Experimental evaluations conducted on a public drivers’ ECG dataset and a custom-built student ECG dataset show that the proposed model effectively generalizes across distinct domains and outperforms state-of-the-art methods in multi-level stress classification. This work is available at https://github.com/Ramyashri14-0593/Stress-Level-Classification.