Aplicação do algoritmo ensemble de floresta aleatória para a classificação de clientes adimplentes e inadimplentes

Abstract

The high default rate leads companies to look for alternatives to estimate delays in their receipts. Default can be largely responsible for a company’s disruption and poor fi-nancial performance. Due its importance, some companies are now investing in the data analysis area, seeking technological advances that make it possible to build data models that can identify the profile of their customers. The objective of this work was to develop a classification model using a random forest to recognize the profile of defaulting or defaulting customers. The research hypothesis is that new opportunities open up when applying the results generated by the model to improve pre-sales, align collection management and adjust cash flow with thelack of defaulting revenue. The proposed model was evaluaeted by using accuracy, precision and recall. In the experiments, we achieve a rate of 95 % for accuracy, 90 % for precision and 0.76 % for recall. The results were promising but additional experiments and evaluations must be carried out before putting the model into production.