Model-Agnostic Explanations with Mimic Rules

Abstract

Recently, a large number of machine learning models have been proposed. Unfortunately, such models often make black-box decisions that are not easy to explain the logical reasons to derive them. Therefore, it is important to develop a tool that automatically gives the reasons for the model’s decision. Some research tackle to solve this problem by approximating an explained model with an interpretable model such as a decision tree. Although these methods provide logical reasons for a model's decision, it sometimes occurs a wrong explanation. We propose a novel model-agnostic explanation method with the rule models that we call mimic rules. Mimic rules are an interpretable model and have the same outputs to an explained model. We give a comparison of our method to previous methods, and we show that our method often improves local fidelity.