Investigating Machine Learning Applications in Credit Risk Assessment

This article presents a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)‑based systematic review of machine learning (ML) applications to corporate credit risk assessment, based on sixty-seven Scopus‑indexed open‑access articles. We identify three research clusters: (i) persistent data challenges, including class imbalance, missing values, and limited reliable financial information; (ii) methodological advances, emphasizing hybrid, ensemble, and optimization‑driven models; and (iii) specialized approaches to small and medium-sized enterprises (SMEs), whose heterogeneity and limited transparency require tailored credit risk analytics. The evidence indicates that ML methods frequently outperform traditional techniques in accuracy, flexibility, and adaptability, while emerging approaches such as explainable AI (XAI), graph‑based models, and quantum‑inspired architectures expand interpretability and predictive depth. Across clusters, two cross‑cutting tensions emerge: between predictive performance and interpretability, and between methodological sophistication and institutional deployability. We propose a conceptual framework that links these tensions to the three clusters and outline a research agenda for translating ML‑based scoring from academic benchmarking to practical adoption in financial institutions.