Personal Credit Scoring for Customers with Limited Credit History: A Hybrid Fuzzy AHP-TOPSIS Model and Evidence from Vietnam

Abstract

Personal credit scoring in developing economies faces a paradox: traditional statistical models require dense historical data, while machine-learning models suffer from limited interpretability. To resolve this paradox, this study proposes and evaluates a Multi-Criteria Decision-Making model combining the Fuzzy Analytic Hierarchy Process (Fuzzy AHP) for criteria weighting with the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) for applicant scoring and classification. Using 340 unsecured consumer-loan files from three Vietnamese commercial banks over 2022-2024, the model is evaluated under limited-data conditions and benchmarked against logistic regression on the same data. The results reveal a trade-off: Fuzzy AHP-TOPSIS attains an Area Under the ROC Curve (AUC) of 0.96 and a bad-debt recall of 0.93, higher than logistic regression (0.93 and 0.79), whereas logistic regression achieves higher precision and F1-score. The proposed model's key strength is operating effectively under such conditions while preserving interpretability. The study recommends a tiered pathway aligned with banks' technological readiness.