Explainable Artificial Intelligence - uma análise dos trade-offs entre desempenho e explicabilidade

Abstract

Explainability is essential for users to efficiently understand, trust, and manage computer systems that use artificial intelligence. Thus, as well as assertiveness, understanding how the decision-making process of the models occurred is fundamental. While there are studies that focus on the explainability of artificial intelligence algorithms, it is important to highlight that, as far as we know, none of them have comprehensively analyzed the trade-offs between performance and explainability. In this sense, this research aims to fill this gap by investigating both transparent algorithms, such as Decision Tree and Logistic Regression, and opaque algorithms, such as Random Forest and Support Vector Machine, in order to evaluate the trade-offs between performance and explainability. The results reveal that opaque algorithms have a low explanability and do not perform well regarding response time due to their complexity, but are more assertive. On the other hand, transparent algorithms have a more effective explainability and better performance regarding response time, but in our experiments, we observed that accuracy obtained was lower than the accuracy of opaque models.