Induction of Non-Monotonic Rules From Statistical Learning Models Using High-Utility Itemset Mining

Making AI decisions more interpretable: The paper demonstrates a method to convert complex statistical learning models (like ensemble trees) into human-readable logic rules, which can help stakeholders understand why an AI system made a particular decision rather than treating it as a "black box."

Significant speed improvements for explainability: The proposed algorithm is substantially faster than existing approaches (ALEPH), which means organizations can extract explanations from AI models more efficiently—an important practical consideration for deploying responsible AI systems at scale.

Leveraging existing explainability tools: The method builds on SHAP/TreeExplainer, established Explainable AI tools, to identify which features matter most in model decisions. This connection to proven interpretability techniques provides a foundation of trust in the extracted explanations.

Better performance without sacrificing interpretability: The experiments show improvements in classification metrics compared to the baseline system, suggesting that creating interpretable rules doesn't necessarily mean accepting worse model performance—a key concern for organizations balancing accuracy with transparency.

Practical applicability to real-world models: By working with ensemble tree models (commonly used in industry), this approach addresses a genuine need: many deployed AI systems use these models, and this method provides a practical pathway to make them more explainable.