Joint Distribution-Informed Shapley Values for Sparse Counterfactual Explanations

Researchers introduce COLA, a framework that refines counterfactual explanations in AI models by using optimal transport theory and Shapley values to achieve the same prediction changes with 26-45% fewer feature modifications. The method works across different datasets and models to create more actionable and clearer AI explanations.

• →COLA framework reduces feature modifications needed for counterfactual explanations by 54-74% while maintaining the same target effects
• →The method uses optimal transport theory to compute coupling between factual and counterfactual data sets
• →Framework is model-agnostic and works as a post-hoc refinement tool for existing counterfactual explanation generators
• →Testing across four datasets, twelve models, and five generators demonstrates broad applicability
• →Approach provides theoretical guarantees that refined counterfactuals don't move farther from originals than initial explanations