- Article
Background: Electroencephalography (EEG) is increasingly used to investigate consumer preferences; however, most predictive studies rely on local spectral or temporal features, and the predictive and statistical relevance of functional connectivity remains underexplored. This study evaluated the statistical significance, predictive performance, and feature-selection stability of EEG functional connectivity measures associated with consumer preference. Methods: Phase Locking Value (PLV) and Normalized Mutual Information (NMI) were computed for the delta, theta, alpha, beta, and gamma frequency bands using product-related EEG recordings from the NeuMa dataset. Subject-level paired statistical analyses were performed with false discovery rate correction. PLV, NMI, and early feature-level PLV+NMI fusion were evaluated using LASSO-based feature selection and five machine-learning classifiers under stratified 10-fold and subject-independent leave-one-subject-out (LOSO) validation. Feature-selection stability was assessed under both validation strategies. Results: Preferred-product viewing was associated with widespread reductions in functional connectivity for both PLV and NMI, with NMI showing particularly extensive differences in the delta band. PLV and NMI yielded broadly comparable predictive performance, with the highest balanced accuracy of 0.699 achieved by PLV combined with Naive Bayes under both stratified 10-fold and LOSO validation. Early PLV+NMI fusion did not improve predictive performance. In contrast, NMI-derived features dominated the fused stability ranking, accounting for 18 of the 20 highest-ranked stable features, predominantly in the delta band. Conclusions: EEG functional connectivity provides complementary information about consumer preference at the statistical, predictive, and feature-stability levels. PLV combined with Naive Bayes provided the highest balanced accuracy, whereas NMI exhibited greater statistical sensitivity and feature-selection stability. These findings highlight low-frequency, particularly delta-band, connectivity as a relevant target for further investigation and demonstrate the importance of evaluating EEG representations beyond classification performance alone.
18 November 2026










