Scholarlink

The detection and interpretation of physiological signals such as electroencephalograms and electrocardiograms have emerged as critical tools in clinical diagnostics for assessing stress and emotional states. This paper introduces AIDAN, a multimodal deep learning system built on a 3D Long Short-Term Memoryarchitecture, designed to analyze temporal patterns in EEG and ECG data, fully developed in MATLAB. AIDAN captures subtle physiological changes over time, enabling reliable classification of stress levels into LOW, MEDIUM, and HIGH categories. The system was evaluated on a small yet representative dataset, achieving notable performance in detecting low-stress levels with an F1 score of 0.58 and a corresponding AUC of 0.37. While medium-stress classification remains challenging due to overlapping signal features, AIDAN demonstrated resilience and adaptability during training, suggesting potential for improvement with enhanced feature representation and class balancing.

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

Deep Learning for Stress Detection – A 3D LSTM Model for EEG and ECG Data Analysis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Amina Radončić

摘要

The detection and interpretation of physiological signals such as electroencephalograms and electrocardiograms have emerged as critical tools in clinical diagnostics for assessing stress and emotional states. This paper introduces AIDAN, a multimodal deep learning system built on a 3D Long Short-Term Memoryarchitecture, designed to analyze temporal patterns in EEG and ECG data, fully developed in MATLAB. AIDAN captures subtle physiological changes over time, enabling reliable classification of stress levels into LOW, MEDIUM, and HIGH categories. The system was evaluated on a small yet representative dataset, achieving notable performance in detecting low-stress levels with an F1 score of 0.58 and a corresponding AUC of 0.37. While medium-stress classification remains challenging due to overlapping signal features, AIDAN demonstrated resilience and adaptability during training, suggesting potential for improvement with enhanced feature representation and class balancing.