Constrained machine learning models for credit default prediction: who wins, and who loses

Aug 3, 2026

Upload by Quinn

Balancing prediction accuracy of machine learning models with interpretability is crucial in credit decision contexts where transparent reasoning is required by regulation. This study examines how monotonicity constraints, used to improve interpretability, influence model performance and credit outcomes in credit default prediction. It compares an XGBoost model with and without monotonicity restrictions using Lending Club data. The analysis identifies which individuals are most affected via changes in predicted default probabilities and evaluates feature-importance shifts using Shapley-regression statistics, addressing bias and fairness implications.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages32

Reading Time1 h 21 min

AccessFree to download

Abstract

Research objective and motivation

Methodology and model comparison

Fairness and interpretability evaluation

RSC 2025/24 Robert Schuman Centre for Advanced Studies Florence School of Banking & Finance Banking Supervision Policy Working Paper Series In the Context of the SSM-EUI Partnership on SSM Banking Supervision Learning Services WORKING PAPER Constrained machine learning models for credit default prediction: who wins, and who loses Andres Alonso-Robisco, Jose Manuel Carbo, Guillermo de Haro, Juan José Guillén García Robert Schuman Centre for Advanced Studies Florence School of Banking & Finance Banking Supervision Policy Working Paper Series In the Context of the SSM-EUI Partnership on SSM Banking Supervision Learning Services Constrained machine learning models for credit default prediction: who wins, and who loses Andres Alonso-Robisco, Jose Manuel Carbo, Guillermo de Haro, Juan José Guillén García RSC Working Paper 2025/24 Banking Supervision Policy Working Paper 2025/05 This work is licensed under the Creative Commons Attribution 4.0 (CC-BY 4.0) International license which governs the terms of access and reuse for this work. If cited or quoted, reference should be made to the full name of the author(s), editor(s), the title, the series and number, the year and the publisher. ISSN 1028-3625 © Andres Alonso-Robisco, Jose Manuel Carbo, Guillermo de Haro, Juan José Guillén García, 2025 Published in June 2025 by the European University Institute. Badia Fiesolana, via dei Roccettini 9 I – 50014 San Domenico di Fiesole (FI) Italy Views expressed in this publication reflect the opinion of individual author(s) and not those of the European University Institute. This publication is available in Open Access in Cadmus, the EUI Research Repository: [https://cadmus.eui.eu](https://cadmus.eui.eu) [www.eui.eu](www.eui.eu) Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Education and Culture Executive Agency (EACEA) . Neither the European Union nor EACEA can be held responsible for them. Robert Schumann Centre for Advanced Studies Robert Schuman Centre for Advanced Studies The Robert Schuman Centre for Advanced Studies, created in 1992 and currently directed by Professor Erik Jones, aims to develop inter-disciplinary and comparative research on the major issues facing the process of European integration, European societies and Europe’s place in 21st century global politics. The Centre is home to a large post-doctoral programme and hosts major research programmes, projects and data sets, in addition to a range of working groups and ad hoc initiatives. The research agenda is organised around a set of core themesand is continuously evolving, reflecting the changing agenda of European integration, the expanding membership of the European Union, developments in Europe’s neighbourhood and the wider world. For more information: [http://eui.eu/rscas](http://eui.eu/rscas) The EUI and the RSC are not responsible for the opinion expressed by the author(s) . Florence School of Banking and Finance The Florence School of Banking and Finance is a key point of reference for training and debate on banking and finance policy at the European level. The aim of the School is to develop a common culture of regulation and supervision in the European Union. It does so through policy debate, training and applied research and in close interaction with its network of leading academic institutions. The banking and financial professionals involved in the School’s activities include members of European and non-European central banks, national supervisory and control authorities, European institutions, academia and the private sector, including numerous high profile figures. The School’s training courses currently focus onfour thematic areas: (1) Financial Regulation and Governance,(2) Digital Finance,(3) Sustainable Finance,(4) Methods. Since its establishment in 2016, more than 24.000

FAQs

Why are interpretability and accuracy balanced in credit default prediction?

Credit decisions require transparent, regulatable decision-making. The document explains that interpretability is critical even when predictive accuracy is the primary goal.

How does the study evaluate monotonicity constraints?

It compares an XGBoost model with and without monotonicity restrictions using Lending Club data, then measures effects on performance and outputs.

Who is most affected by the monotonicity constraints according to the findings?

The study focuses on individuals whose predicted default probabilities change the most and assesses related shifts in feature importance using Shapley-regression-based analysis to consider bias and fairness.

Aug 3, 2026

Upload by Quinn