The Application of Personal Credit Scoring Models in Consumer Credit

Economics and Finance, University of Bristol, Bristol, United Kingdom

^* Corresponding author: qw24015@bristol.ac.uk

Abstract

Traditional techniques like logistic regression and decision trees are appreciated due to their interpretability, regulatory acceptance, and minimal implementation expenses. But they primarily depend on formal data and are not good at scoring “thin-file” borrowers or handling non-traditional data sources, so they are less helpful in diverse markets. By contrast, artificial intelligence-based models — like XGBoost, LSTM networks, and digital footprint models—are providing greater predictive power and flexibility by accessing unstructured data and identifying complex behavioral patterns. Despite such strengths, AI systems also present challenges, including lack of interpretability, data bias, fairness, and high resource intensity. The report also discusses the key regulatory and ethical considerations with comparisons across jurisdictions and references case studies like Indonesia’s fintech sector to demonstrate both the promise and perils of AI-driven scoring. The review finishes by highlighting the necessity for well-balanced, understandable, and fair AI solutions supported by strong data governance and cross-disciplinary cooperation.