Beyond Plausibility: Actionable Counterfactual Explanations via Causal Intervention Discovery

Abstract

Counterfactual explanations (CFEs) are increasingly used to explain machine learning model predictions by identifying the smallest changes to an input that would flip the prediction. However, many current CFE methods focus primarily on generating plausible counterfactuals without explicitly considering whether those counterfactuals are actionable in the real world. This paper addresses this limitation by introducing a novel framework for generating actionable CFEs based on causal intervention discovery. Our approach leverages causal models to identify interventions on modifiable variables that are most likely to lead to the desired outcome, while also respecting causal dependencies and constraints. We formulate the problem as a causal optimization problem and propose an efficient algorithm to find actionable CFEs. We evaluate our approach on several benchmark datasets and demonstrate that it generates CFEs that are significantly more actionable and causally sound compared to existing methods, without sacrificing plausibility or proximity. The results highlight the importance of incorporating causal reasoning into CFE generation to improve the practical utility and trustworthiness of explanations.