Search papers, labs, and topics across Lattice.
This paper introduces Fast Evidential Rule Learning (FERL), a novel method for interpretable classification that integrates evidential reasoning directly into fuzzy rule models, allowing for abstention when the model is uncertain. FERL achieves statistically significant improvements in accuracy over existing rule learners across a benchmark of 30 tabular datasets, with an average accuracy increase of 2.6% compared to the next best method. Additionally, FERL demonstrates superior performance in out-of-distribution detection and attribute anomaly identification, showcasing its robustness and practical applicability in real-world scenarios.
FERL achieves a 2.6% accuracy boost over state-of-the-art rule learners while providing interpretable, abstention-capable outputs directly from a single inference pass.
Interpretable classification often requires more than accurate predictions for real-life deployment: models should be transparent about the evidence behind their decisions and abstain when they cannot decide reliably. We introduce Fast Evidential Rule Learning (FERL), a method that learns interpretable, accurate fuzzy rule models whose outputs are evidential. Unlike post-hoc calibration, FERL's belief, plausibility, and abstention capabilities arise directly from the fuzzy memberships in a single deterministic pass, with no auxiliary head, held-out set, or repeated inference. Our theoretical analysis further shows that FERL is Lipschitz stable, which means that its evidential outputs vary smoothly with the input. Against state-of-the-art rule learners, FERL is statistically significantly more accurate across a 30 tabular-dataset benchmark ($+2.6\%$ average accuracy over the second best). Its native set predictions attain the best utility-discounted accuracy among credal classifiers ($u_{65}/u_{80}=0.80/0.83$ vs.\ $0.79/0.80$ for the naive credal classifier), at higher set coverage ($0.92$ vs.\ $\le0.82$). FERL also matches dedicated out-of-distribution detectors on tabular near-OOD detection ($77.7$ vs.\ $77.4$ AUROC for the strongest baseline). Under detector-class-disjoint concept-bottleneck evaluation, its it is within $2.3$ AUROC points of the strongest dedicated detector on both CUB and AwA2, while attaining the best AwA2 AUPR-Out ($68.3$) and novel-class rejection ($57.2$), while being able to name which attributes are anomalous.
