Scholarlink

Using multi-modal signals like Electroencephalogram (EEG) and Electrocardiogram (ECG) offers a powerful approach to emotion recognition as it takes input from both central and peripheral physiological responses. This study explores multi-modal emotion and stress recognition using EEG and ECG signals from the benchmark dataset, DREAMER. Signal features such as entropy and fractal dimension were extracted from both EEG and ECG signals. Additionally, Power Spectral Density (PSD) features from EEG signals and Heart Rate Variability (HRV) features from ECG signals were obtained. A pipeline involving data pre-processing, feature extraction, and model training for classification was implemented. For a comparative analysis of efficiency, models such as Random Forest, Support Vector Machine, Gradient Boosting, XGBoost, LightGBM, Voting Classifier, and Stacking Classifier were used. The Stacking Classifier demonstrated the best overall performance, achieving a test accuracy of 92% and an average cross-validation score of 89% during 10-fold cross-validation for stress classification. Additionally, it achieved test accuracies of 90% for valence, 98% for arousal, and 99% for dominance. Explanations for predictions using Shapley Additive exPlanations (SHAP) values inferred that entropy and HRV characteristics are consistently recognized as crucial indicators, confirming their reliability as dependable biomarkers for emotion and stress detection across a wide variety of human subjects.

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

Multimodal Emotion and Stress Recognition Using EEG and ECG: A Machine Learning Approach with Explainable AI

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • K. L. Amritha Nandini,
  • Anerud Thiyagarajan,
  • S. Vishal,
  • R. Giri Prasath,
  • Amrutha Veluppal

摘要

Using multi-modal signals like Electroencephalogram (EEG) and Electrocardiogram (ECG) offers a powerful approach to emotion recognition as it takes input from both central and peripheral physiological responses. This study explores multi-modal emotion and stress recognition using EEG and ECG signals from the benchmark dataset, DREAMER. Signal features such as entropy and fractal dimension were extracted from both EEG and ECG signals. Additionally, Power Spectral Density (PSD) features from EEG signals and Heart Rate Variability (HRV) features from ECG signals were obtained. A pipeline involving data pre-processing, feature extraction, and model training for classification was implemented. For a comparative analysis of efficiency, models such as Random Forest, Support Vector Machine, Gradient Boosting, XGBoost, LightGBM, Voting Classifier, and Stacking Classifier were used. The Stacking Classifier demonstrated the best overall performance, achieving a test accuracy of 92% and an average cross-validation score of 89% during 10-fold cross-validation for stress classification. Additionally, it achieved test accuracies of 90% for valence, 98% for arousal, and 99% for dominance. Explanations for predictions using Shapley Additive exPlanations (SHAP) values inferred that entropy and HRV characteristics are consistently recognized as crucial indicators, confirming their reliability as dependable biomarkers for emotion and stress detection across a wide variety of human subjects.