Scholarlink

<p>Timely and precise identification of stress is critical for mental health assessment, yet remains a challenging task. Traditional methods, such as self-reported questionnaires and physiological cues, often suffer from subjectivity, inconsistency, and limited reliability. EEG-based approaches offer a more objective alternative, but they also face their own challenges, including signal noise, complex spatial–temporal dependencies, and a lack of robust uncertainty modeling. This article presents a novel deep learning framework for early stress detection, called U-AGGDCNet + SAO, to address these issues. The framework integrates multiple stages: EEG denoising through the iterative robust peak-aware guided filter, feature extraction using the multiple discrete orthonormal S-transform, classification via an uncertainty-aware gated graph decision capsule network, and performance optimization with snow ablation optimization. This hybrid design enhances signal clarity, captures discriminative neural patterns, and improves learning stability in the face of uncertainty. Significantly lower error rates, achieving up to 99.9% accuracy and 98.8% F1-score, are demonstrated through experimental validation on the SAM40 and DEAP datasets. The reliability, robustness, and generalizability of the proposed framework are confirmed by these results, making it a viable option for early-stage stress monitoring with real-world EEG assistance.</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

Early Stress Detection and Analysis Using EEG Signals-Based Uncertainty-Aware Gated Graph Decision Capsule Network with Snow Ablation Optimization

  • 期刊论文

  • 发表日期：

    2025年12月11日

查看全文
  • Ashvini Bamanikar,
  • Ritesh V. Patil,
  • Lalit V. Patil,
  • Surendra Mahajan

摘要

Timely and precise identification of stress is critical for mental health assessment, yet remains a challenging task. Traditional methods, such as self-reported questionnaires and physiological cues, often suffer from subjectivity, inconsistency, and limited reliability. EEG-based approaches offer a more objective alternative, but they also face their own challenges, including signal noise, complex spatial–temporal dependencies, and a lack of robust uncertainty modeling. This article presents a novel deep learning framework for early stress detection, called U-AGGDCNet + SAO, to address these issues. The framework integrates multiple stages: EEG denoising through the iterative robust peak-aware guided filter, feature extraction using the multiple discrete orthonormal S-transform, classification via an uncertainty-aware gated graph decision capsule network, and performance optimization with snow ablation optimization. This hybrid design enhances signal clarity, captures discriminative neural patterns, and improves learning stability in the face of uncertainty. Significantly lower error rates, achieving up to 99.9% accuracy and 98.8% F1-score, are demonstrated through experimental validation on the SAM40 and DEAP datasets. The reliability, robustness, and generalizability of the proposed framework are confirmed by these results, making it a viable option for early-stage stress monitoring with real-world EEG assistance.