Obtaining classification rules using LVQ +PSO: An application to credit risk

Abstract

Credit risk management is a key element of financial corporations. One of the main problems that face credit risk officials is to approve or deny a credit petition. The usual decision making process consists in gathering personal and financial information about the borrower. This paper present a new method that is able to generate classifying rules that work no only on numerical attributes, but also on nominal attributes. This method, called LVQ+PSO, combines a competitive neural network with an optimization technique in order to find a reduced set of classifying rules. These rules constitute a predictive model for credit risk approval. Given the reduced quantity of rules, our method is very useful for credit officers aiming to make decisions about granting a credit. Our method was applied to two credit databases that were extensively analyzed by other competing classification methods. We obtain very satisfactory results. Future research lines are exposed.