Optimization-based Counterfactual Analysis for Explainable Decision-Making
This research develops an optimization based perspective on counterfactual analysis to improve the interpretability of data driven decision making systems. Counterfactual explanations are typically constructed locally, often overlooking how modeling choices shape the resulting optimization problems.
We propose a mathematical programming framework in which counterfactual analysis is formulated as optimization models minimizing perturbation costs under predictive constraints. The framework incorporates asymmetric costs and sparsity constraints and accommodates group settings with multiple instances under explicit allocation rules.
These modeling principles also support more complex settings, including functional data and decision making problems where the outcome is itself the solution of an optimization model, enabling interpretable and actionable explanations.
Keywords: Counterfactual Explanations Explainability Mathematical optimization