Credit classification using regulation techniques on the Credit German database

The development of microfinance, as well as microcredit, has generated greater competition among financial institutions to attract customers in this business segment. For this reason, the development of credit scoring models is highly required by these financial institutions. In this sense, to ensure that no overfitting is generated in the use of prediction techniques and in case of difficulty with collinearity, it will not be possible to obtain reliable estimates and predictions through common statistical techniques such as least squares; for this reason, it is significant and necessary to apply regularized regression methods such as Ridge, Lasso and Elastic Net. The present research determined the optimal credit scoring model for the Credit German database using the Ridge, Lasso, and Elastic Net regulation techniques. This dataset was initially analyzed with the Logit model, finding that this model has an accuracy of 37.2%; on the other hand, the Lasso model presented an accuracy of 76.7%, the Ridge model of 75.6%, and the Elastic Net model of 69.2%. Finally, the Lasso model evidenced the best prediction of the credit rating of Credit German future clients, with an accuracy in the training data of 82.9% and for the test data of 76.7%, being superior to the proposed models.