Related Experiment Video
Updated: Nov 23, 2025
08:00
Decoding Natural Behavior from Neuroethological Embedding
Published on: October 3, 2025
311
Deep Neural Networks for Behavioral Credit Rating
Andro Merćep1, Lovre Mrčela1, Matija Birov2
1Laboratory for Financial and Risk Analytics, Faculty of Electrical Engineering and Computing, University of Zagreb, 10000 Zagreb, Croatia.
Entropy (Basel, Switzerland)
|December 30, 2020
Summary
This study introduces a deep neural network for behavioral credit rating, outperforming traditional models. This advanced approach enhances credit risk prediction accuracy for banks and consumers, meeting regulatory demands.
Area of Science:
- * Computational finance
- * Machine learning applications in banking
Background:
- * Logistic regression is the standard for credit risk modeling.
- * Regulatory demands for explainability hinder advanced machine learning adoption.
- * Non-linear algorithms, like deep neural networks, offer superior prediction accuracy.
Purpose of the Study:
- * Propose a deep neural network for behavioral credit rating.
- * Enhance accuracy in assessing existing loan portfolio performance.
- * Meet Basel regulatory framework capital requirements.
Main Methods:
- * Developed and trained a deep neural network model.
- * Utilized two datasets: 2009-2013 (financial crisis) and 2014-2018 (post-crisis).
- * Data comprised over 1.5 million loan examples.
Main Results:
- * The deep neural network surpassed multiple benchmark models.
- * Performance was comparable to the XGBoost model.
- * Analyzed long-term credit rating performance and impact of reprogrammed facilities.
Conclusions:
- * Deep neural networks offer a viable, high-performance alternative to logistic regression in credit risk.
- * The proposed model effectively addresses regulatory requirements for explainability and accuracy.
- * Advanced models can benefit both consumers and financial institutions.
