Deep Learning para avaliação de risco de crédito financeiro

Abstract

Banks and financial organizations are dealing with credit lending as the key to success. The credit assessment models rank their prospective clients as likely defaulters or not. Credit risk assessment is often evaluated by a manager from his or her intuitive experience. However, it is possible to make these substantial decisions more accurately through the support of the models developed from machine learning and data mining, using essentially the classification task. Several data mining techniques, such as decision trees, random forests, support vector machines, neural networks, have already been used to classify prospective customers. The objective of this work is the evaluation of consolidated classificatory algorithms in the literature and the suggestion of the construction of a deep neural networks model applied to the problem of credit risk assessment. The experiments were conducted on three publicly available data sets in the UCI Machine Learning Repository. The average accuracy of the techniques was compared. The results obtained show that the random forest classifier excels to the other techniques because it presents the highest average accuracy and is the most stable for the three databases studied. The technique of deep learning presented a good result but did not obtain a result superior to the one mentioned previously.