Explaining the black-box or using the black-box to develop better interpretable solutions?

Authors

  • Vinicius Alves Matias Universidade de São Paulo
  • Julia Machado Lechi Universidade de São Paulo
  • Norton Trevisan Roman Universidade de São Paulo
  • Luciano Antonio Digiampietri Universidade de São Paulo

DOI:

https://doi.org/10.5335/rbca.v17i2.16459

Keywords:

Black Box Models, Explainable AI, Interpretable Models, Post-hoc Interpretability

Abstract

Understanding the decision-making processes behind Artificial Intelligence models became a crucial aspect of AI. This paper describes a study that compares the performance of models produced by both interpretable and black-box algorithms and evaluates if it is possible to use black-box models to assist in interpretable models' training. We verified a significant difference in performance between the two types of models. However, the interpretable model was able to mimic the behavior of the black-box models to a satisfactory degree. The promising initial results obtained from using black-box models to aid in interpretable models' training suggest the potential efficacy of this approach.

Downloads

Download data is not yet available.

Downloads

Published

2025-08-16

Issue

Section

Original Paper

How to Cite

[1]
2025. Explaining the black-box or using the black-box to develop better interpretable solutions?. Brazilian Journal of Applied Computing. 17, 2 (Aug. 2025), 46–53. DOI:https://doi.org/10.5335/rbca.v17i2.16459.