A Systematic Review of AI-Driven Credit Risk Assessment Models in Commercial Banking (2018–2026)

Abstract

This systematic review examines the evolution, methodological advancements, and governance implications of AI-driven credit risk assessment models in commercial banking from 2018 to 2026. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, the review synthesizes evidence from 27 peer-reviewed studies, representing a diverse body of research exploring machine learning and deep learning applications in probability of default estimation, loss and exposure modeling, early-warning systems, portfolio monitoring, and automated credit decisioning. The findings show that AI models consistently outperform traditional statistical approaches in predictive accuracy and behavioral sensitivity, particularly when leveraging ensemble architectures and temporal or transactional features. However, despite these technical advantages, the review identifies substantial constraints related to data quality, model generalizability, explainability, fairness, drift vulnerability, and regulatory acceptability. Many studies highlight persistent challenges in achieving transparency and stability under stress scenarios, indicating that current AI systems often struggle to meet prudential and consumer-protection expectations. The review also notes a widening gap between research innovation and real-world deployment, as operational requirements such as continuous monitoring, documentation, and interoperability create significant barriers to adoption. Overall, this study provides a comprehensive evidence base demonstrating both the promise and limitations of AI credit risk models, emphasizing the need for more interpretable model architectures, standardized validation frameworks, privacy-preserving data ecosystems, and cross-institutional benchmarking. These insights contribute to shaping the next generation of trustworthy, governable, and regulator-aligned AI systems for commercial banking.