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

Trust in automation is critical for effective human-automation interaction, yet traditional subjective measures are limited in capturing rapid and dynamic changes in user trust. This study introduces PLVNet, a novel deep neural network architecture designed to classify trust versus distrust states from EEG functional connectivity features derived using Phase Locking Value (PLV). PLV features were extracted across six canonical EEG frequency bands (Delta, Theta, Alpha, Beta, Low Gamma, High Gamma) from 30-channel EEG recordings. The PLVNet model was evaluated using three complementary approaches: aggregated analysis (5× 5 stratified cross-validation), participant-wise analysis, and leave-one-subject-out (LOSO) cross-validation. PLVNet significantly outperformed convolutional neural network (CNN), support vector machine (SVM) and k-nearest neighbours (KNN) classifiers across all evaluation schemes. Beta and Low Gamma bands provided the highest discriminative power, while functional connectivity analysis revealed that trust is associated with enhanced fronto-parietal and fronto-occipital synchronisation, reflecting global network integration, whereas distrust shows fragmented connectivity patterns. PLVNet's ability to capture non-linear inter-dependencies in connectivity patterns highlights its advantages over conventional methods. These findings demonstrate that PLV-based connectivity robustly reflects trust-related neural dynamics, underscoring the potential of PLVNet for real-time, objective monitoring of trust in human-automation systems, which paves the way for adaptive and neuro-aware interfaces.