Machine learning XAI for early loan default prediction
Abstract: Early default prediction with predictive models is of crucial importance for financial institutions, Fintech or Peer to Peer (P2P) lending platforms, as it allows them to effectively mitigate the potential risks associated with customer or debtor defaults, anticipating before this becomes...
| Autores: | , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universidad Complutense de Madrid (UCM) |
| Repositorio: | Docta Complutense |
| Idioma: | inglés |
| OAI Identifier: | oai:docta.ucm.es:20.500.14352/132721 |
| Acceso en línea: | https://hdl.handle.net/20.500.14352/132721 |
| Access Level: | acceso abierto |
| Palabra clave: | 004.85 519.2 336.77 Peer to peer lending Machine learning Early default risk XAI Fuzzy Inteligencia artificial (Informática) Estadística 1203.04 Inteligencia Artificial 1209 Estadística 1209.14 Técnicas de Predicción Estadística |
| Sumario: | Abstract: Early default prediction with predictive models is of crucial importance for financial institutions, Fintech or Peer to Peer (P2P) lending platforms, as it allows them to effectively mitigate the potential risks associated with customer or debtor defaults, anticipating before this becomes a major problem. This proactive approach serves to avoid the consequent impact on provisions and, subsequently, on the institution's capital. On the other hand, advanced predictive models are often less interpretable than traditional models such as probit (Abdou & Pointon, 2011) and logistic regression (Bolton, 2009; Liu et al. 2024). Due to this lower explainability, our goal was to develop a methodology that allows building an advanced predictive model together with a linguistically interpretable explanation useful for decision making from large volumes of data. For this purpose, our case study was the loan dataset of Lending Club, the largest P2P lending platform in the world. As a result, we obtained a model based on the eXtreme Gradient Boosting (XGBoost) together with its linguistic interpretation using a surrogate model and the 2-tuple fuzzy linguistic model Monje et al., (Mathematics 10:1428, 2022). This model allows us to identify five risk categories (very low, low, medium, high and very high). |
|---|