Explainability and business sense in machine learning models for credit risk assesment
Loading...
Download
Official URL
Full text at PDC
Publication date
2024
Defense date
18/12/2023
Authors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Complutense de Madrid
Citation
Abstract
El cambiante panorama de la modelización del riesgo de crédito, debido al avance tecnológico y la proliferante generación de plataformas con alternativas de financiación, alternas alas instituciones tradicionales, ha planteado cuestiones críticas en relación con la eficacia del aprendizaje automático para satisfacer las exigencias de los reguladores y de los usuarios, en particular en el mercado de préstamos P2P. En este contexto, se abordan algunos vacíos identificados en la literatura en relación con la modelización, su aplicabilidad orientada al negocio y la interpretabilidad de los resultados del aprendizaje automático. Proponemos estrategias alternativas de modelización, flujos de trabajo, incorporando predictibilidad sentido de negocio y explicabilidad. Se demuestra que este enfoque no sólo mejora la precisión de las predicciones frente a alternativas tradicionales, sino que permite identificar las variables determinantes, y a su vez, perfiles de riesgo o segmentos de mayor rentabilidad...
The changing landscape of credit risk modeling, due to technological advancement and the proliferating generation of platforms with alternative financing alternatives to traditional institutions, has raised critical questions regarding the effectiveness of machine learning in meeting the demands of regulators and users, particularly in the P2P lending market. In this context, we address some gaps identified in the literature regarding modeling, its business-oriented applicability, and the interpretability of machine learning results. We propose alternative modeling strategies workflows, incorporating business sense predictability and explainability. We show that this approach not only improves the accuracy of predictions compared to traditional alternatives but also allows the identification of the critical variables and, in turn, risk profiles or segments of higher profitability...
The changing landscape of credit risk modeling, due to technological advancement and the proliferating generation of platforms with alternative financing alternatives to traditional institutions, has raised critical questions regarding the effectiveness of machine learning in meeting the demands of regulators and users, particularly in the P2P lending market. In this context, we address some gaps identified in the literature regarding modeling, its business-oriented applicability, and the interpretability of machine learning results. We propose alternative modeling strategies workflows, incorporating business sense predictability and explainability. We show that this approach not only improves the accuracy of predictions compared to traditional alternatives but also allows the identification of the critical variables and, in turn, risk profiles or segments of higher profitability...
Description
Tesis inédita de la Universidad Complutense de Madrid, Facultad de Estudios Estadísticos, leída el 18-12-2023. Tesis formato europeo (compendio de artículos)