ExMo: Explainable AI Model using Inverse Frequency Decision Rules

ExMo uses simple IF...THEN rules for transparency: The model makes decisions based on straightforward conditional statements (e.g., "IF feature X is present THEN predict fraud"), which can be directly explained to users without requiring technical expertise to understand why a prediction was made.

Better rule selection improves accuracy: By using TF-IDF (a technique that identifies which features are most distinctive for each category), ExMo extracts more relevant decision rules than previous methods like Bayesian Rule Lists, resulting in 20% higher accuracy on fraud detection tasks.

Achieves deep learning-level performance with human-readable explanations: ExMo matches the accuracy of complex deep learning models while remaining interpretable, addressing a key challenge in responsible AI where organizations often must choose between accuracy and explainability.

Designed for non-expert understanding: The paper emphasizes that ExMo provides "human-friendly" explanations, suggesting the approach prioritizes accessibility for business users and stakeholders who need to understand and trust AI decisions but lack machine learning backgrounds.

Validated on real-world applications: Testing on actual fraud detection scenarios demonstrates practical applicability beyond academic benchmarks, indicating the method could be deployed in high-stakes decision-making contexts where explainability is legally or ethically required.