Case-based Explainability for Random Forest: Prototypes, Critics, Counter-factuals and Semi-factuals

Case-based explanations make AI decisions more understandable by showing actual examples: Rather than using abstract rules, this approach explains why a Random Forest model made a decision by pointing to real data examples—such as similar past cases (prototypes), contrasting cases (critics), or "what-if" scenarios (counter-factuals)—making the reasoning more concrete and relatable to users.

This method is particularly valuable for regulated industries like finance: The paper emphasizes that explainability through case-based reasoning addresses regulatory requirements and the need for transparency in financial and other heavily regulated sectors, where decision-makers must be able to justify why an AI system approved or denied a loan, for example.

Random Forests can now be explained using their own internal logic: By extracting the distance metric that Random Forests naturally learn during training, researchers can identify meaningful examples that align with how the model actually makes decisions, rather than imposing external explanation methods that may not reflect the model's true reasoning.

Multiple types of explanations serve different purposes: The paper identifies four complementary explanation types—prototypes (similar supporting cases), critics (similar opposing cases), counter-factuals (minimal changes needed to flip a decision), and semi-factuals (partial changes that shift the decision)—allowing stakeholders to understand decisions from multiple angles.

These explanations can be systematically evaluated for quality: The research uses various metrics to assess whether the identified examples actually provide meaningful explanatory power, helping organizations verify that their explanations are genuinely informative rather than just plausible-sounding.