fbpx

A two-stage fuzzy neural approach for credit risk assessment in a Brazilian credit card company

Tipo
Artigos

Ano
08/07/2020

Linha de Pesquisa
Administração e Economia de Negócios

Autor(es)
Diego Paganoti Fonseca, Peter Wanke, Henrique Luiz Correa

Orientador

https://doi.org/10.1016/j.asoc.2020.106329


Applied Soft Computing, v. 92. Abstract: This study explores and evaluates the use of soft computing systems for clients’ credit risk assessment in a Brazilian private credit card provider through the development of an innovative two-stage process, both involving soft computing techniques (fuzzy and neural networks). We use commercially available credit score ratings both in the development of our method and for benchmarking. After describing the development of our method, we present a discussion about the comparison of performances of our method and a number of other credit scoring methods described in literature (for e.g. statistical and soft computing-based). One of the analyzed existing methods for instance involves the use of a soft computing algorithm only – Artificial Neural Networks (ANN) – for client classification into solvent or non-solvent, having a market available credit score rating as input. One of the most relevant contributions of this study however is the development of what we consider an innovative approach for credit scoring. This is a two-stage process that involves the use of a fuzzy inference model as input for an ANN model (what we call a fuzzy-neural approach), using commercially available credit score ratings as response in order to conduct the fuzzy reasoning step of the analysis. The main conclusion of our research is that, in general, our fuzzy-neural method had better results than the pure application of some market available score rating method as input to a Multi-Layer Perceptron (MLP) since it was able to reduce uncertainty by improving predictability and reducing variability of the outcomes when compared to a model with no scores. The performance of a combination of a fuzzy and a neural method was very satisfactory; the vagueness usually present in the information of a company’ database was to a certain extent, incorporated by our method with good results. From the practical perspective, although our method has not proved to be substantially better than market available options, we demonstrated that it is possible for companies to develop credit score rating mechanisms internally based on past data by using fuzzy inference systems. Under certain circumstances, companies may find this option preferable than the usual option of paying high fees to large credit scoring agencies for the use of their proprietary systems.

Rolar para cima