Associating Deep Learning and Logistic Regression for Credit Card Default Prediction

Título: Associating Deep Learning and Logistic Regression for Credit Card Default Prediction

Autores: Marcus Vinicius de Oliveira, Gilson Costa

Resumo: This paper proposes an hybrid approach that integrates Deep Learning models with logistic regression to enhance the prediction of credit card defaults while maintaining the explainability of final classification using a partially explainable methodology. We compare the performance of this proposed methodology with conventional classification techniques as well as a pure Multilayer Perceptron model. The primary goal of our architecture is to improve the interpretability of the classification results, enabling users to critically evaluate the algorithm’s predictions and identify the variables that have the most significant impact. This understanding aims to facilitate the development of more effective business strategies.

Palavras-chave: Credit Default; Machine Learning; Classification; Interpretability.

Páginas: 8

Código DOI: 10.21528/CBIC2025-1191056

Artigo em PDF: CBIC_2025_paper1191056.pdf

Arquivo BibTeX:
CBIC_2025_1191056.bib