Optimal Counterfactual Explanations for Scorecard modelling
This work addresses the need for explainable AI in banking lending decisions, but it is incremental as it focuses on a specific interpretable model type rather than broad black-box models.
The authors tackled the problem of generating counterfactual explanations for scorecard models in banking, proposing mixed-integer programming formulations that produce close, realistic, and sparse counterfactuals with diversity, and experiments on real-world datasets showed it can generate optimal diverse counterfactuals with reasonable CPU times.
Counterfactual explanations is one of the post-hoc methods used to provide explainability to machine learning models that have been attracting attention in recent years. Most examples in the literature, address the problem of generating post-hoc explanations for black-box machine learning models after the rejection of a loan application. In contrast, in this work, we investigate mathematical programming formulations for scorecard models, a type of interpretable model predominant within the banking industry for lending. The proposed mixed-integer programming formulations combine objective functions to ensure close, realistic and sparse counterfactuals using multi-objective optimization techniques for a binary, probability or continuous outcome. Moreover, we extend these formulations to generate multiple optimal counterfactuals simultaneously while guaranteeing diversity. Experiments on two real-world datasets confirm that the presented approach can generate optimal diverse counterfactuals addressing desired properties with assumable CPU times for practice use.