Aplicação de análise de risco de crédito com o uso das técnicas de regressão logística e árvores de decisão

Abstract

With the increase in credit granting and the need for a quick and assertive response to requests for this type of service, it is necessary to use methods for measuring credit risk in order to make the decision process more reliable. This work aims to compare and exemplify the use of two techniques for this purpose, namely: logistic regression and decision tree using the CART method. For this, a database related to the granting of a credit card by a certain banking institution was selected. The database of interest was divided into two subsets of training (adjustment) and validation (verification), the latter being twenty percent of the size of the initial database. The predictions obtained by adjusting both methods were compared in the training and validation data through performance indicators, such as sensitivity, specificity and area under the ROC curve. The results were satisfactory and quite similar in both cases, however the logistic regression outperformed the CART decision tree in our application.