Izvorni znanstveni članak
https://doi.org/https://doi.org/10.2478/bsrj-2026-0017
Credit Risk Assessment in Banks Using Machine Learning: Insights from Industry Practice and a Case Study
Branka Tuškan Sjauš
; University of Zagreb, Faculty of Economics and Business, Croatia
Ana Ivanišević Hernaus
; University of Zagreb, Faculty of Economics and Business, Croatia
Gabriela Paradžik
; University of Zagreb, Faculty of Economics and Business, Croatia
Sažetak
Background: The study investigates differences between traditional and machine learning (ML) methods for credit risk assessment in banking, highlighting ML's role and performance as an emerging tool. Objectives: The paper explores credit risk assessment in banking practice and presents a case study of an ML tool in credit analysis to identify its effectiveness, challenges, and role in promoting financial inclusion compared with logistic regression. Methods/Approach: We used a mixed-methods approach. The focus was qualitative, using semi-structured interviews with representatives from Croatia's banking sector regulator and the five largest banks. We also applied quantitative analysis to the German Credit Dataset, using logistic regression as a traditional model and a decision tree as an ML model. Results: Expert interviews emphasized the importance of regulatory compliance and effective employee training in ML credit analysis. In quantitative analysis, logistic regression proved more accurate and easier to interpret, though it struggled to capture nonlinear patterns– potentially leading to the exclusion of some applicants¬–where the decision tree proved more successful. Conclusions: The study confirms that ML has potential in credit risk assessment because it can process extensive data quickly and effectively. However, it still has drawbacks to address, including interpretability, overfitting, accuracy, data quality, and regulatory compliance. The paper's findings may serve policymakers, as well as professionals and researchers in this field.
Ključne riječi
credit risk assessment; machine learning; logistic regression; decision tree; financial inclusion
Hrčak ID:
351423
URI
Datum izdavanja:
24.9.2026.
Posjeta: 0 *