Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improving Compactness and Reducing Ambiguity of CFIRE Rule-Based Explanations

Sebastian Müller, Tobias Schneider, Ruben Kemna, Vanessa Toborek

TL;DR

The paper addresses inter-class ambiguity in CFIRE's rule-based explanations for tabular data by introducing ThresholdPruning, a post-hoc procedure that uses per-term win counts $Wins[t]$ and a tolerance $\theta$ to trim the global rule set $E$ while controlling relative accuracy on a reference set $X$. Safe pruning with $\theta=0$ preserves predictive performance while reducing both model size and cross-class conflicts, and modest pruning with $\theta=0.05$ achieves additional gains with only minor F1 loss (typically $<2\%$). Across 700 networks and 14 tasks using three local explainers, pruning yields large reductions in ambiguity and compactness, particularly for larger CFIRE models, with some tasks remaining inherently ambiguous. These findings support deploying CFIRE with a pruning extension to produce clearer, more compact explanations without sacrificing accuracy.

Abstract

Models trained on tabular data are widely used in sensitive domains, increasing the demand for explanation methods to meet transparency needs. CFIRE is a recent algorithm in this domain that constructs compact surrogate rule models from local explanations. While effective, CFIRE may assign rules associated with different classes to the same sample, introducing ambiguity. We investigate this ambiguity and propose a post-hoc pruning strategy that removes rules with low contribution or conflicting coverage, yielding smaller and less ambiguous models while preserving fidelity. Experiments across multiple datasets confirm these improvements with minimal impact on predictive performance.

Improving Compactness and Reducing Ambiguity of CFIRE Rule-Based Explanations

TL;DR

The paper addresses inter-class ambiguity in CFIRE's rule-based explanations for tabular data by introducing ThresholdPruning, a post-hoc procedure that uses per-term win counts and a tolerance to trim the global rule set while controlling relative accuracy on a reference set . Safe pruning with preserves predictive performance while reducing both model size and cross-class conflicts, and modest pruning with achieves additional gains with only minor F1 loss (typically ). Across 700 networks and 14 tasks using three local explainers, pruning yields large reductions in ambiguity and compactness, particularly for larger CFIRE models, with some tasks remaining inherently ambiguous. These findings support deploying CFIRE with a pruning extension to produce clearer, more compact explanations without sacrificing accuracy.

Abstract

Models trained on tabular data are widely used in sensitive domains, increasing the demand for explanation methods to meet transparency needs. CFIRE is a recent algorithm in this domain that constructs compact surrogate rule models from local explanations. While effective, CFIRE may assign rules associated with different classes to the same sample, introducing ambiguity. We investigate this ambiguity and propose a post-hoc pruning strategy that removes rules with low contribution or conflicting coverage, yielding smaller and less ambiguous models while preserving fidelity. Experiments across multiple datasets confirm these improvements with minimal impact on predictive performance.
Paper Structure (6 sections, 2 figures, 2 tables, 1 algorithm)

This paper contains 6 sections, 2 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background: CFIRE
  3. Rule Pruning
  4. Experiments
  5. Results
  6. Conclusion

Figures (2)

  • Figure 1: Schematic overview of the original CFIRE pipeline with the proposed ThresholdPruning (Algorithm \ref{['alg:pruning']}) extension.
  • Figure 2: Trade-off between F1 and relative effects of pruning on Size and Ambiguity. Color encodes dataset; markers circle, square and triangle show means over $\Phi$s for no pruning, safe pruning $\theta=0$ and threshold pruning $\theta=0.05$.