Many model explanations focus on answering the question: Why did the model make this prediction Counterfactual explanations ask a different question: What would need to change for the prediction to be different? Instead of analyzing contributions or sensitivities, counterfactual methods look for alternative scenarios.
Given a specific individual and a specific outcome, for example, a rejected loan application, we ask: what is the smallest change that would have resulted in approval? A counterfactual explanation might say:
- If income were slightly higher,
- Or if debt were slightly lower,
- Or if both changed moderately,
the model would have produced a different decision. This makes counterfactual explanations:
- Action-oriented
- Individualized
- Often intuitive
They focus on recourse: how outcomes could change.
