Scholarlink

In this study, a multimodal classifier has been developed to analyze the spectral power of EEG signals, trained on data representing three states: wakefulness, fatigue, and drowsiness. The classifier demonstrated high performance, achieving an accuracy of 91.67% and f1-score of 0.94. These results confirmed its potential application for automatic real-time labeling of data, enabling the rapid generation of an extensive dataset for further analysis of psychophysiological states based on ECG. For each subject, the parameters of the most significant heart rate variability (HRV) correlating to the error rates during monotonous activity were identified. The highly correlated parameters are HRV-Prc80NN, HRV-pNN50, HRV-pNN20 and HRV-MedianNN. The results of the ECG signal classification using the selected parameters showed a significant dependence on the characteristics of the individual subjects. Among subjects, the highest classification accuracy reached 86.4%, while the lowest was 65.1%, using the Random Forest model. Thus, the proposed approach demonstrates significant potential for automatic labeling of EEG data and for the classification of ECG data in the context of identifying wakefulness and fatigue states.

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 Machine Learning Approach to Classification of Operator’s Psychophysiological States During Monotonous Activity Based on EEG and ECG Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Denis V. Kostulin,
  • Pavel D. Shaposhnikov,
  • Avedik Kh. Ekizyan,
  • Mikhail G. Shevchenko,
  • Dmitry G. Shaposhnikov

摘要

In this study, a multimodal classifier has been developed to analyze the spectral power of EEG signals, trained on data representing three states: wakefulness, fatigue, and drowsiness. The classifier demonstrated high performance, achieving an accuracy of 91.67% and f1-score of 0.94. These results confirmed its potential application for automatic real-time labeling of data, enabling the rapid generation of an extensive dataset for further analysis of psychophysiological states based on ECG. For each subject, the parameters of the most significant heart rate variability (HRV) correlating to the error rates during monotonous activity were identified. The highly correlated parameters are HRV-Prc80NN, HRV-pNN50, HRV-pNN20 and HRV-MedianNN. The results of the ECG signal classification using the selected parameters showed a significant dependence on the characteristics of the individual subjects. Among subjects, the highest classification accuracy reached 86.4%, while the lowest was 65.1%, using the Random Forest model. Thus, the proposed approach demonstrates significant potential for automatic labeling of EEG data and for the classification of ECG data in the context of identifying wakefulness and fatigue states.