Local Universal Explainer (LUX) -- a rule-based explainer with factual, counterfactual and visual explanations

LUX provides multiple explanation types in one tool: Unlike many existing explainers that focus on single aspects, LUX generates factual explanations (what the model actually decided), counterfactual explanations (what would need to change for a different decision), and visual explanations all together, offering a more complete picture of how an AI system reached its decision.

The method uses real data clusters rather than generating artificial data: LUX prioritizes identifying high-density clusters of actual data points that influence the model's decision boundary, reducing reliance on artificially generated samples. This approach may produce more trustworthy explanations grounded in real-world patterns rather than synthetic scenarios.

LUX outperforms comparable explainers on key metrics: In testing against established rule-based explainers (LORE, EXPLAN, and Anchor), LUX demonstrated superior performance in simplicity, fidelity (accuracy of explanations), representativeness (how well explanations reflect actual data), and consistency (reliability across similar cases).

The tool integrates with existing AI explanation methods: By incorporating SHAP (a widely-used feature importance technique) through a novel sampling algorithm, LUX builds on established XAI approaches rather than requiring entirely new infrastructure, potentially making it easier for organizations to adopt.

Rule-based explanations offer interpretability: By using modified decision tree algorithms to generate rule-based explanations, LUX produces outputs that are inherently more human-readable than black-box model outputs, supporting the goal of making AI decisions understandable to non-technical stakeholders.