설명 가능한 인공지능(XAI)을 활용한 금융 기회 불평등의 진단: 교차성(Intersectionality) 공정성을 중심으로 = Diagnosis of Financial Opportunity Inequality using XAI: Focusing on Intersectional Fairness

다국어 초록 (Multilingual Abstract)

As the adoption of AI in the financial sector accelerates, algorithmic bias has become a critical issue. This study diagnoses the fairness of AI models from the perspective of providing 'financial opportunity,' moving beyond the traditional focus on risk management. We constructed a VIP prediction model using Random Forest based on credit card usage behavior data and analyzed latent biases using SHAP (Shapley Additive Explanations) and a proposed 'Intersectional Stress Testing' method based on counterfactual scenarios. The results showed that while the model appeared rational in univariate analysis (e.g., gender alone), intersectional analysis combining gender, age, and credit history revealed a 'Systemic Exclusion' phenomenon, with a 75.5-fold gap in VIP approval probability between disadvantaged and advantaged groups. This study empirically demonstrates that AI can amplify complex discrimination in limited information environments and suggests that fairness verification must expand from single-attribute analysis to multivariate intersectional diagnosis.