There are now many explainable AI methods for understanding the decisions of a machine learning model. Among these are those based on counterfactual reasoning, which involve simulating features changes and observing the impact on the prediction. This article proposes to view this simulation process as a source of creating a certain amount of knowledge that can be stored to be used, later, in different ways. This process is illustrated in the additive model and, more specifically, in the case of the naive Bayes classifier, whose interesting properties for this purpose are shown.

Counterfactual explanations have become a mainstay of the XAI field. This particularly intuitive statement allows the user to understand what small but necessary changes would have to be made to a given situation in order to change a model prediction. The quality of a counterfactual depends on several criteria: realism, actionability, validity, robustness, etc. In this paper, we are interested in the notion of robustness of a counterfactual. More precisely, we focus on robustness to counterfactual input changes. This form of robustness is particularly challenging as it involves a trade-off between the robustness of the counterfactual and the proximity with the example to explain. We propose a new framework, CROCO, that generates robust counterfactuals while managing effectively this trade-off, and guarantees the user a minimal robustness. An empirical evaluation on tabular datasets confirms the relevance and effectiveness of our approach.

Improvements of machine learning techniques for decision systems has led to the rise of applications in various domains such as healthcare, credit or justice. The eventual sensitivity of such domains, as well as the black-box nature of the algorithms, has motivated the need for methods that explain why some prediction was made. For example, if a person's loan is rejected as a result of a model decision, the bank must be able to explain why. In such a context, it might be interesting to provide an explanation of what that person should change to influence the model's decision. As suggested by Wachter et al. [27], one way to build this type of explanation is through the use of counterfactual explanations. A counterfactual is defined as the smallest modification of feature values that changes the prediction of a model to a given output. In addition, the explanation also provides important feedback to the user. In the context of a denied credit, a counterfactual is a close individual for whom his credit is accepted and the feature modifications suggested by a counterfactual acts as recourse for the user. For privacy reason or simply because there is no similar individual with an opposite decision, we aim to generate synthetic individuals as counterfactuals.