Abstract
Motor imagery-based brain-computer interfaces (MI-BCIs) remain limited by inter-subject variability, nonstationary EEG characteristics, and the difficulty of identifying a reliable golden subject for subject-to-subject transfer learning. Previous version of HAFL study used fuzzy decision by opinion score method (FDOSM) to identify a golden subject. However, the decision-making layer can be strengthened by explicitly separating criteria weighting from compromise ranking. This study extends the published framework by developing a unified 9 × 18 Golden Subject Decision Matrix (UGSDM), where nine subjects are evaluated simultaneously across nine training-phase criteria and nine evaluation-phase criteria. FWZIC is used to calculate the importance weights of the eighteen criteria, and VIKOR is used to obtain the final compromise ranking of the nine subject alternatives. The FWZIC results show that the highest weights were assigned to key reliability and performance criteria, including classification accuracy, recall/sensitivity, F1-score, LogLoss, specificity for both training and evaluation phase criteria. The VIKOR results ranked S1 as the best compromise subject with Q = 0.0000, followed by S3 with Q = 0.1722 and S5 with Q = 0.3373. This result differs from the previous FDOSM-based ranking, which identified S5 as the golden subject. The proposed FWZIC-VIKOR framework provides a transparent and reusable decision-making layer for golden subject benchmarking. The ranking shift from S5 to S1 shows that the new weighted compromise-ranking logic can reveal different subject priorities when training and evaluation criteria are combined into a single decision matrix. Sensitivity analysis across four weighting scenarios confirmed the robustness of the proposed framework, with (S1) ranked first in three scenarios and second in the training-dominant scenario.