Fast Evidential Rule Learning for Interpretable Classification
August 7, 2026
FERL enables interpretable fuzzy rule models that provide inherent belief, plausibility, and abstention capabilities in a single deterministic pass. It outperforms state-of-the-art rule learners with a 2.6% average accuracy increase across 30 tabular datasets.
HOW THIS AFFECTS YOU
●
builderYou can deploy more reliable tabular models that natively know when to abstain from making a prediction.
●
researcherThis method provides a Lipschitz stable way to achieve evidential outputs without auxiliary heads or extra inference steps.