A HYBRID NEURO-FUZZY MODEL FOR ASSESSING BANK BORROWER CREDITWORTHINESS
Main Article Content
Abstract
Introduction. In the context of growing macroeconomic instability, digital transformation, and increasing uncertainty in financial markets, the problem of ensuring the reliability of lending decisions and minimizing credit risk has become particularly relevant. Traditional approaches to assessing borrower creditworthiness are largely based on statistical and scoring methods that require significant volumes of historical data and often fail to account for the uncertainty, nonlinearity, and qualitative characteristics inherent in borrowers’ behavior. Under such conditions, the application of intelligent decision-support methods based on fuzzy logic and artificial intelligence technologies becomes increasingly important.
Purpose. The purpose of this study is to analyze the peculiarities of applying fuzzy approaches to borrower creditworthiness assessment and to develop a hybrid neuro-fuzzy model based on the Mamdani algorithm and the Adaptive Neuro-Fuzzy Inference System (ANFIS). The proposed model is aimed at improving the accuracy and interpretability of credit risk assessment by integrating quantitative and qualitative borrower characteristics into a unified decision-support framework.
Results. The study proposes a two-level hybrid neuro-fuzzy model for assessing bank borrower creditworthiness. At the first level, individual indicators characterizing the borrower's social status, financial condition, debt servicing quality, and lending conditions are aggregated into corresponding sub-indices using the Mamdani fuzzy inference algorithm. At the second level, the obtained sub-indices are integrated into a single creditworthiness indicator using ANFIS, which enables the automatic adjustment of membership functions and identification of nonlinear dependencies between factors.
The proposed model is characterized by a hierarchical structure and a high degree of adaptability, allowing the composition of input variables to be modified depending on the type of loan product, including mortgage, consumer, and automobile lending. The architecture of the model was tested using synthetically generated datasets comprising 500 conditional borrowers. The results demonstrated that the implementation of ANFIS at the aggregation stage reduced the root mean square error by 42.6% and decreased the mean absolute percentage error from 11.4% to 6.2% compared with the classical expert-based Mamdani system.
The developed approach allows simultaneous consideration of quantitative and qualitative borrower characteristics, eliminates rigid threshold limitations of traditional scoring methods, and provides a transparent interpretation of the obtained results. In addition, the model is capable of identifying compensation effects among individual factors and adapting to changes in the external economic environment.
Conclusions. The findings confirm the effectiveness of applying hybrid neuro-fuzzy approaches to borrower creditworthiness assessment under conditions of uncertainty and incomplete information. The proposed model combines the interpretability of expert fuzzy systems with the predictive capabilities of adaptive neuro-fuzzy technologies, thereby improving the quality of credit decision-making and enhancing the efficiency of bank credit risk management systems. The practical implementation of the developed model can contribute to reducing the share of non-performing loans and increasing the resilience of banking institutions to macroeconomic shocks. Future research should focus on validating the model using real banking data and comparing its performance with modern machine learning methods, including XGBoost, Random Forest, and deep neural networks.
Article Details
This work is licensed under a Creative Commons Attribution 4.0 International License.
The authors published in this journal agree with following conditions:
1. The authors reserve to themselves the right to the authorship of their works and transfer the right of their first publication to the journal on the terms of Creatіve Common Attrіbutіon Lіcense which allows to freely extend to other persons the published work with an obligatory reference to the authors of the original work and its first publication in this journal.
2. The authors have the right to complete independent additional agreements concerning the not exclusive distribution of their work in the form in which it was published in this journal (for example, to place the work in the electronic storehouse of an establishment or to publish as a monograph component), under the condition of the preservation of the reference to the first publication of the work in this journal.
3. The journal’s policy allows and encourage the authors to place their manuscripts into the Internet (for example, in depositories of establishments or on personal web-sites) either before submitting of the manuscript for publication or during its editorial processing as it assists the occurrence of a productive scientific discussion and positively affects the efficiency and dynamics of citing of the published work.
AGREEMENT
ABOUT TRANSMISSION OF COPYRIGHT
I, the author of the article / We, the authors of the manuscript _______________________________________________________________________
in case of its acceptance for publication, we transfer the following rights to the founders and editorial boards of the scientific publication "BULLETIN OF THE CHERKASY BOHDAN KHMELNYTSKY NATIONAL UNIVERSITY. ECONOMIC SCIENCES. SERIES "ECONOMIC SCIENCES":
1. Publication of this article in Ukrainian (English, Russian, Polish) and distribution of its printed version.
2. Dissemination of the electronic version of the article through any electronic means (placing on the official journal web site, in electronic databases, repositories, etc.).
At the same time we reserve the right without consent of the editorial board and the founders:
1. Use the materials of the article in whole or in part for educational purposes.
2. To use the materials of the article in whole or in part for writing your own theses.
3. Use article materials to prepare abstracts, conference reports, and oral presentations.
4. Post electronic copies of the article (including the final electronic version downloaded from the journal's official website) to:
a. personal web-pecypcax of all authors (web sites, web pages, blogs, etc.);
b. web-pecypcax of the institutions where the authors work (including electronic institutional repositories);
with. non-profit, open-source web-pecypcax (such as arXiv.org).
With this agreement, we also certify that the submitted manuscript meets the following criteria:
1. Does not contain calls for violence, incitement of racial or ethnic enmity, which are disturbing, threatening, shameful, libelous, cruel, indecent, vulgar, etc.
2. Does not infringe the copyrights and intellectual property rights of others or organizations; contains all the references to the cited authors and / or publications envisaged by applicable copyright law, as well as the results and facts used in the article by other authors or organizations.
3. It has not been previously published in other publishers and has not been published in other publications.
4. Does not include materials that are not subject to publication in the open press, in accordance with applicable law.
____________________ ___________________
First name, Last name, signature of the author
"___" __________ 20__
References
Zadeh, L. A. (1965). Fuzzy Sets. Information and Control, 8(3), 338–353. Retrieved from: https://doi.org/10.1016/S0019-9958(65)90241-X (Accessed 21.11.2025).
Jang, J.-S. R. (1993). ANFIS: Adaptive-Network-Based Fuzzy Inference System. IEEE Transactions on Systems, Man, and Cybernetics, 23(3), 665–685. Retrieved from: https://doi.org/10.1109/21.256541 (Accessed 21.11.2025).
Shovhun, N. V. (2012). Analysis of borrower creditworthiness using fuzzy logic methods. Visnyk NTUU «KPI». Informatyka, upravlinnia ta obchysliuvalna tekhnika, 55, 169–173. Retrieved from: https://ela.kpi.ua/handle/123456789/3713 (Accessed 21.11.2025) [in Ukrainian].
Novoseletskyi, O. M., & Yakubets, O. V. (2013). Assessment of individual borrower creditworthiness using a fuzzy-set approach. Naukovi zapysky Natsionalnoho universytetu «Ostrozka akademiia». Seriia «Ekonomika», 24, 242–246. Retrieved from: https://journals.oa.edu.ua/Economy/article/view/763/621 (Accessed 21.11.2025) [in Ukrainian].
Markovych, T. H. (2019). Development of express analysis of exporter enterprises’ creditworthiness based on the fuzzy logic method. Problemy teorii ta metodolohii bukhhalterskoho obliku, kontroliu i analizu, 2(43), 55–60. Retrieved from: https://doi.org/10.26642/pbo-2019-2(43)-55-60 (Accessed 21.11.2025) [in Ukrainian].
Bennouna, G., & Tkiouat, M. (2018). Fuzzy logic approach applied to credit scoring for microfinance in Morocco. Procedia Computer Science, 127, 274–283. Retrieved from: https://doi.org/10.1016/j.procs.2018.01.123 (Accessed 21.11.2025).
Sanchez-Roger, M., Oliver-Alfonso, M. D., & Sanchis-Pedregosa, C. (2019). Fuzzy Logic and Its Uses in Finance: A Systematic Review Exploring Its Potential to Deal with Banking Crises. Mathematics, 7(11), Article 1091. Retrieved from: https://doi.org/10.3390/math7111091 (Accessed 21.11.2025).
Vu, H. (2025). Fuzzy Logic in Credit Risk Assessment: A Literature Review (Bachelor's thesis). Espoo: Aalto University. Retrieved from: https://aaltodoc.aalto.fi/items/08918f5b-3036-49ae-9549-33b90efa7df6 (Accessed 21.11.2025).
Thomas, L. C., Crook, J. N., & Edelman, D. B. (2017). Credit Scoring and Its Applications. Philadelphia: SIAM. Retrieved from: https://doi.org/10.1137/1.9781611974560 (Accessed 21.11.2025).
Louzada, F., Ara, A., & Fernandes, G. B. (2016). Classification methods applied to credit scoring: Systematic review and overall comparison. Surveys in Operations Research and Management Science, 21(2), 117–134. Retrieved from: https://doi.org/10.1016/j.sorms.2016.09.001 (Accessed 21.11.2025).
Basel Committee on Banking Supervision. (2006). Sound Credit Risk Assessment and Valuation for Loans. Basel: Bank for International Settlements. Retrieved from: https://www.bis.org/publ/bcbs126.htm (Accessed 21.11.2025).
MathWorks. (2024). Adaptive Neuro-Fuzzy Inference System (ANFIS). Natick, MA: The MathWorks, Inc. Retrieved from: https://www.mathworks.com/help/fuzzy/adaptive-neuro-fuzzy-inference-system.htm (Accessed 21.11.2025).