HanCred: Heterogeneous graph attention networks for credit risk assessment in a Super-App environment
Financial credit risk scoring is of the utmost importance for credit lenders, a well-designed risk management strategy allows lenders to make better decisions in terms of collocation of financial products and credit limits while keeping losses bounded. In this paper, we propose a Heterogeneous Graph...
- Autores:
-
Acevedo Viloria, Jaime David
- Tipo de recurso:
- Fecha de publicación:
- 2021
- Institución:
- Universidad de los Andes
- Repositorio:
- Séneca: repositorio Uniandes
- Idioma:
- spa
- OAI Identifier:
- oai:repositorio.uniandes.edu.co:1992/55695
- Acceso en línea:
- http://hdl.handle.net/1992/55695
- Palabra clave:
- Graph Neural Networks
Credit Risk
Super-App
Alternative Data
Ingeniería
- Rights
- openAccess
- License
- https://repositorio.uniandes.edu.co/static/pdf/aceptacion_uso_es.pdf
| Summary: | Financial credit risk scoring is of the utmost importance for credit lenders, a well-designed risk management strategy allows lenders to make better decisions in terms of collocation of financial products and credit limits while keeping losses bounded. In this paper, we propose a Heterogeneous Graph Attention Network model that leverages the many interactions found in a Super-App for the Credit Risk assessment of its users with only Alternative Data. Through the proposed model we manage to learn not only for the scarce amount of users with credit history but also from the users with no credit history; promoting financial inclusion in the process. We test our model in a real-life database of 3.5k users drawn from a Super-App, proving that our model achieves an AUC of 0.69 in terms of classifying whether the users will pay back or not. |
|---|