Classificação de Solicitações de Crédito com Algoritmos Supervisionados: Uma Abordagem com SVM e Árvores de Decisão

Abstract

This work proposes the use of Machine Learning techniques to analyze credit applications from clients of a risk management institution. Two approaches were developed: one using Decision Tree and the other using Support Vector Machine (SVM). Using a real dataset with behavioral and financial attributes, the objective was to predict whether a credit application would be approved or rejected. The results indicate better performance for the Decision Tree (accuracy 79.3%, recall 0.83) compared to the SVM (accuracy 73.2%, recall 0.57), with emphasis on its interpretability. The study also analyzed the impact of removing redundant features and the interpretability capacity of the models. Both approaches proved to be effective, with the Decision Tree standing out for its interpretability and the SVM for its stability.