Assessing cognitive mental workload via EEG signals and an ensemble deep learning classifier based on denoising...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Operator performance in human-machine collaboration is crucial.
• Accurate estimation of cognitive states, specifically mental workload (MW), is essential for reliability.
• Existing methods for MW estimation using electroencephalogram (EEG) data have limitations in personalization and handling inter-subject variability.

Purpose of the Study:

• To propose a novel human mental workload recognizer using deep learning and EEG features.
• To develop a method that accounts for personalized cognitive properties and reduces variations between subjects.
• To enhance the reliability and accuracy of operator performance estimation in collaborative environments.

Main Methods:

• Utilized deep learning principles, specifically a stacked denoising autoencoder (SDAE) with a feature mapping layer for preserving local EEG dynamics.
• Introduced a subject-specific integrated deep learning committee for an ensemble classifier.
• Developed the ensemble SDAE classifier with local information preservation (EL-SDAE) for EEG data analysis.

Main Results:

• The EL-SDAE classifier demonstrated strong performance in recognizing human mental workload.
• The proposed method effectively preserved local information in EEG dynamics, enabling personalized feature mapping.
• The ensemble classifier adapted to individual cognitive properties, alleviating inter-subject feature variations.
• EL-SDAE outperformed several classical mental workload estimators in classification accuracy.

Conclusions:

• The EL-SDAE model offers a reliable approach for estimating human mental workload using EEG.
• This deep learning-based method enhances operator performance assessment in human-machine collaborative settings.
• The personalization capability of EL-SDAE addresses key challenges in inter-subject EEG data analysis.