PLVNet: EEG-based trust classification using Phase Locking Value connectivity and deep neural networks.

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Machine Learning

Background:

• Effective human-automation interaction relies on accurate user trust assessment.
• Traditional subjective trust measures are insufficient for dynamic, real-time evaluation.
• Objective neurophysiological markers are needed to capture rapid trust fluctuations.

Purpose of the Study:

• Introduce PLVNet, a deep neural network for classifying trust states using EEG functional connectivity.
• Evaluate PLVNet's performance against traditional machine learning classifiers.
• Investigate neural correlates of trust and distrust in human-automation systems.

Main Methods:

• Extracted Phase Locking Value (PLV) functional connectivity features from 30-channel EEG across six frequency bands.
• Developed and evaluated PLVNet, a novel deep neural network architecture.
• Employed aggregated, participant-wise, and leave-one-subject-out cross-validation for robust assessment.

Main Results:

• PLVNet significantly outperformed CNN, SVM, and KNN classifiers in trust state classification.
• The Beta and Low Gamma frequency bands showed the highest discriminative power.
• Trust correlated with enhanced fronto-parietal/occipital synchronisation, indicating global integration; distrust showed fragmented connectivity.

Conclusions:

• PLVNet effectively captures non-linear inter-dependencies in EEG connectivity for trust monitoring.
• PLV-based connectivity robustly reflects neural dynamics associated with user trust.
• PLVNet offers a pathway for real-time, objective trust assessment in adaptive human-automation systems.