Ai for Credit Risk Modeling: a Deep Learning Approach
Year: 2025
Type: article
Abstract: Credit risk modeling is a critical component of financial decision-making, enabling lenders to assess the probability of default and optimize credit allocation. Traditional credit scoring models, including logistic regression and decision tree-based classifiers, have limitations in handling non-linear financial relationships, class imbalance, and borrower heterogeneity. Recent advances in artificial intelligence (AI) and deep learning (DL) have introduced more sophisticated models capable of capturing complex borrower patterns while improving risk assessment accuracy. However, AI-driven credit risk models must address key challenges, including class imbalance, fairness, model interpretability, and scalability in real-world financial environments. This study proposes a comprehensive DL-based credit risk modeling framework that integrates graph neural networks (GNNs), generative adversarial networks (GANs), and adversarial fairness learning to enhance credit risk prediction accuracy, fairness, and adaptability across borrower segments. The model leverages autoencoders for feature extraction, cost-sensitive learning for imbalanced classification, and domain adaptation techniques for improved model robustness. Additionally, an explainability layer is incorporated to enhance transparency in credit decision-making. Experiments on real-world credit datasets demonstrate that the proposed framework outperforms traditional credit risk models, achieving higher recall for defaulters, reduced bias in loan approvals, and improved computational efficiency. The findings highlight the potential of AI-driven credit risk modeling to transform risk assessment strategies, ensuring more accurate, fair, and scalable credit allocation for financial institutions.
DOI: 10.61784/mjet3023
Authors: Kaoru Fujisawa, Renji Takeda, Aiko Mori
Institutions: Nagoya Institute of Technology, Tohoku University
Language: English
Subfield: Accounting
Domain: Social Sciences
SDG: Reduced inequalities
Open Access status: hybrid
OpenAlex ID: W4408464222