Análise de Risco de Crédito para Gestoras de Vendedores Porta-a-Porta usando Aprendizado de Máquina

  • Ricardo Zorzal Davila Programa de Pós-Graduaçãoo em Computação Aplicada, Instituto Federal do Espírito Santo (IFES) - Campus Serra, ES / Máximus Soluções
  • Francisco de Assis Boldt Programa de Pós-Graduaçãoo em Computação Aplicada, Instituto Federal do Espírito Santo (IFES) - Campus Serra, ES
  • Filipe Mutz Programa de Pós-Graduaçãoo em Computação Aplicada, Instituto Federal do Espírito Santo (IFES) - Campus Serra, ES
Keywords: Credit Risk Analysis, Credit Scoring, Artificial Intelligence, Machine Learning, XGBoost, Neural Networks, Logistic Regression

Abstract

Credit risk analysis is fundamental for small and medium companies given that defaults cause significant impacts in revenue. Banks and other financial institutions have access to several information regarding the financial wellbeing of clients for credit analysis. On the other hand, companies from other areas need to estimate credit risk using demographic information and the history of interations with the company. This work studies the problem of credit risk analysis for companies that manage door-to-door salesman. These companies have specific characteristics that differenciate them from businesses of other niches. These specificities influence the process of credit risk analysis. A dataset was built using information from partner companies e this dataset was used to train several machine learning algorithms in the task of default prediction. Experiments showed that the logistic regression classifiers achieves the same performance as nonlinear classifiers (e.g., neural networks and XGBoost), while being less prone to overfitting and being more interpretable.
Published
2021-10-20
Section
Articles