Scholarlink

<p>Psychological stress has become a critical public health concern, underscoring the need for accurate and scalable detection methods. Electroencephalography (EEG), with its non-invasive nature and high temporal resolution, offers a direct window into brain dynamics and holds significant potential for stress monitoring. In a proposed machine learning (ML) approach to the problem of stress classification using EEG signals, the researchers suggested an integrated approach that derives and instils into the proposed framework custom signal preprocessing, frequency band decomposition, and dimensionality reduction. EEG recordings in DASPS were separated into Theta, Beta and Alpha bands. To handle the high dimensionality and the noise of the signal in feature extraction, Principal Component Analysis and Fast Independent Component Analysis were worked on. Five ML classifiers, Support Vector Machine, Linear Discriminant Analysis, RF, AdaBoost, and Gradient Boosting, were evaluated using key metrics such as Accuracy, Precision, Recall, and ROC. Gradient Boosting consistently achieved superior results, with a peak accuracy of 97.20% using PCA on Alpha band features. Additionally, Exploratory Data Analysis revealed interpretable neurophysiological patterns linked to stress. The proposed framework demonstrates strong generalizability and reliability, offering a robust foundation for real-time, non-invasive mental health monitoring systems using EEG signals.</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

Stress detection using EEG signals: comparative analysis of machine learning models and feature extraction

  • 期刊论文

  • 发表日期：

    2025年10月30日

查看全文
  • Goldy Saini,
  • Rakesh Kumar,
  • Lokesh Malviya,
  • Arvind Mewada,
  • Mohd. Aquib Ansari

摘要

Psychological stress has become a critical public health concern, underscoring the need for accurate and scalable detection methods. Electroencephalography (EEG), with its non-invasive nature and high temporal resolution, offers a direct window into brain dynamics and holds significant potential for stress monitoring. In a proposed machine learning (ML) approach to the problem of stress classification using EEG signals, the researchers suggested an integrated approach that derives and instils into the proposed framework custom signal preprocessing, frequency band decomposition, and dimensionality reduction. EEG recordings in DASPS were separated into Theta, Beta and Alpha bands. To handle the high dimensionality and the noise of the signal in feature extraction, Principal Component Analysis and Fast Independent Component Analysis were worked on. Five ML classifiers, Support Vector Machine, Linear Discriminant Analysis, RF, AdaBoost, and Gradient Boosting, were evaluated using key metrics such as Accuracy, Precision, Recall, and ROC. Gradient Boosting consistently achieved superior results, with a peak accuracy of 97.20% using PCA on Alpha band features. Additionally, Exploratory Data Analysis revealed interpretable neurophysiological patterns linked to stress. The proposed framework demonstrates strong generalizability and reliability, offering a robust foundation for real-time, non-invasive mental health monitoring systems using EEG signals.