Cognitive Load Prediction From Multimodal Physiological Signals Using Multiview Learning.

Area of Science:

• Human-Computer Interaction
• Neuroscience
• Physiological Computing

Background:

• Accurate cognitive load prediction is vital for human-computer interaction, especially in high-stakes environments like aviation.
• Existing multimodal fusion methods require adaptation for robust cognitive load classification.

Purpose of the Study:

• To propose a feature selection framework using multiview learning to improve cognitive load prediction.
• To address information redundancy and identify common physiological mechanisms of cognitive load.

Main Methods:

• A feature selection-multiview classification with cohesion and diversity (FS-MCCD) framework was developed.
• Multimodal physiological signals (EEG, EDA, ECG, EOG, eye movements) were collected during Multi-Attribute Task Battery (MATB) tasks.
• Feature selection integrated view and feature weights to optimize the prediction model.

Main Results:

• The cognitive load prediction model achieved an average accuracy of 81.08% and an F1-score of 80.94% for three-class classification.
• Feature analysis revealed links between high cognitive load and specific EEG patterns (increased delta/theta, decreased alpha) and increased pupil diameter.
• The framework effectively fused multimodal features for efficient cognitive load prediction.

Conclusions:

• The proposed FS-MCCD framework demonstrates effectiveness and efficiency in predicting cognitive load using multimodal physiological signals.
• Understanding the physiological underpinnings of cognitive load can inform the design of adaptive human-computer systems.
• This approach offers a promising avenue for real-time cognitive state monitoring in critical applications.