Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fuzzy Implicative Rules: A Unified Approach

Raquel Fernandez-Peralta

TL;DR

This work reframes rule mining by modeling fuzzy rules as logical implicatives, introducing the monotonicity of the generalized modus ponens ($MTC$) as a key property to ensure coherent rule evaluation and monotone support. It identifies adequate operator pairs $(T,I)$ that satisfy both ($TC$) and ($MTC$), including several families of fuzzy implications and t-norms, and shows how the framework generalizes crisp and other fuzzy-rule perspectives. The authors provide an open-source Python implementation for mining fuzzy implicative associative rules and demonstrate directional pattern discovery and model similarity across datasets, highlighting the importance of operator choice for capturing directional information. The approach offers a flexible, interpretable framework that can augment traditional rule mining and XAI applications, with future work on broader operator design and integration into additional rule-mining tasks.

Abstract

Rule mining algorithms are one of the fundamental techniques in data mining for disclosing significant patterns in terms of linguistic rules expressed in natural language. In this paper, we revisit the concept of fuzzy implicative rule to provide a solid theoretical framework for any fuzzy rule mining algorithm interested in capturing patterns in terms of logical conditionals rather than the co-occurrence of antecedent and consequent. In particular, we study which properties should satisfy the fuzzy operators to ensure a coherent behavior of different quality measures. As a consequence of this study, we introduce a new property of fuzzy implication functions related to a monotone behavior of the generalized modus ponens for which we provide different valid solutions. Also, we prove that our modeling generalizes others if an adequate choice of the fuzzy implication function is made, so it can be seen as an unifying framework. Further, we provide an open-source implementation in Python for mining fuzzy implicative associative rules. We test the applicability and relevance of our framework for different real datasets and fuzzy operators.

Fuzzy Implicative Rules: A Unified Approach

TL;DR

This work reframes rule mining by modeling fuzzy rules as logical implicatives, introducing the monotonicity of the generalized modus ponens () as a key property to ensure coherent rule evaluation and monotone support. It identifies adequate operator pairs that satisfy both () and (), including several families of fuzzy implications and t-norms, and shows how the framework generalizes crisp and other fuzzy-rule perspectives. The authors provide an open-source Python implementation for mining fuzzy implicative associative rules and demonstrate directional pattern discovery and model similarity across datasets, highlighting the importance of operator choice for capturing directional information. The approach offers a flexible, interpretable framework that can augment traditional rule mining and XAI applications, with future work on broader operator design and integration into additional rule-mining tasks.

Abstract

Rule mining algorithms are one of the fundamental techniques in data mining for disclosing significant patterns in terms of linguistic rules expressed in natural language. In this paper, we revisit the concept of fuzzy implicative rule to provide a solid theoretical framework for any fuzzy rule mining algorithm interested in capturing patterns in terms of logical conditionals rather than the co-occurrence of antecedent and consequent. In particular, we study which properties should satisfy the fuzzy operators to ensure a coherent behavior of different quality measures. As a consequence of this study, we introduce a new property of fuzzy implication functions related to a monotone behavior of the generalized modus ponens for which we provide different valid solutions. Also, we prove that our modeling generalizes others if an adequate choice of the fuzzy implication function is made, so it can be seen as an unifying framework. Further, we provide an open-source implementation in Python for mining fuzzy implicative associative rules. We test the applicability and relevance of our framework for different real datasets and fuzzy operators.

Paper Structure

This paper contains 11 sections, 14 theorems, 33 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Fuzzy implicative rules and quality measures
  4. Rule's Modeling
  5. Fuzzy implicative rules as a generalization of others
  6. Quality measures
  7. Solutions for adequate fuzzy operators
  8. Experiments
  9. Mining directional patterns
  10. Similarity between models
  11. Conclusions

Key Result

Proposition 7

Let $C$ be a copula. The function $I_C: [0,1]^2 \to [0,1]$ given by is a fuzzy implication function if and only if for all $x_1 \leq x_2 \text{ and } y \in [0,1]$. In this case, $I_C$ is called a probabilistic implication.

Figures (4)

  • Figure 1: Synthetic dataset illustrating an induced asymmetric dependency between two variables $A$ and $B$.
  • Figure 2: Evolution of support and confidence for the rules $A$ High $\to$$B$ High and $B$ High $\to$$A$ High as a function of the parameters $\lambda$ and $p$ of the pairs $(T_{\lambda}^{\bm{SS}},I^{\bm{SS}}_{k_{\lambda}})$ and $(T_{\bm{LK}}\xspace,I_p)$, respectively.
  • Figure 3: Effect of minimum fuzzy coverage on rule quality and set size (top 20% of rules by confidence). Left axis: mean confidence and mean support; right axis: number of rules. Fuzzy operators were fixed to $(I_{\bm{LK}}\xspace,I_p)$ with $p=0.01$.
  • Figure 4: Mean similarity between the set of rules obtained in Table \ref{['table:results1']} for different models.

Theorems & Definitions (40)

  • Definition 1: Klement2000
  • Example 2
  • Definition 3: Baczynski2008
  • Example 4
  • Definition 5
  • Definition 6: Klir1995
  • Proposition 7: Baczynski2016
  • Definition 8: Zhou2021
  • Definition 9
  • Proposition 10
  • ...and 30 more