XAI LIMETool: Uma ferramenta que avalia explicabilidade em Inteligência Artificial

Abstract

This paper proposes a tool to evaluate the quality of explanations generated by the LIME algorithm, widely used to increase the interpretability of opaque machine learning models. The tool applies the faithfulness metric, which verifies if the explanations truly reflect the behavior of the original model, reinforcing transparency and reliability in predictions. The system was tested with the Pima Indians Diabetes Database and the Random Forest model. Based on this, local explanations of predictions were generated, allowing a detailed analysis of how the attributes influence the model's decisions.