Explainable AI for Classification using Probabilistic Logic Inference

Explainability can match accuracy: The paper demonstrates that classification systems designed to be transparent and explainable can achieve performance levels comparable to traditional "black box" methods like random forests and neural networks, challenging the assumption that transparency requires sacrificing accuracy.

Symbolic knowledge bases enable human understanding: By converting training data into a symbolic Knowledge Base and using probabilistic logic, the method creates a system where decisions can be traced through explicit logical rules rather than hidden mathematical patterns, making it easier for humans to understand why a classification was made.

Identifies which features matter most: The approach pinpoints the specific features (data inputs) that are decisive for each classification decision, providing explanations similar to SHAP, a leading method in the field—this helps users understand which information the system actually relied upon.

Bridges symbolic and probabilistic reasoning: The method combines symbolic logic (rule-based reasoning) with probabilistic inference (handling uncertainty), offering a middle ground between purely rule-based systems that are rigid and purely statistical systems that are opaque.

Practical performance across diverse datasets: Testing on both synthetic and real-world datasets shows the approach is robust and generalizable, suggesting it could be practically applied to real classification problems where both accuracy and explainability matter.