Redundancy, Deduction Schemes, and Minimum-Size Bases for Association Rules

Association rules are among the most widely employed data analysis methods in the field of Data Mining. An association rule is a form of partial implication between two sets of binary variables. In the most common approach, association rules are parameterized by a lower bound on their confidence, which is the empirical conditional probability of their consequent given the antecedent, and/or by some other parameter bounds such as “support” or deviation from independence. We study here notions of redundancy among association rules from a fundamental perspective. We see each transaction in a dataset as an interpretation (or model) in the propositional logic sense, and consider existing notions of redundancy, that is, of logical entailment, among association rules, of the form “any dataset in which this first rule holds must obey also that second rule, therefore the second is redundant”. We discuss several existing alternative definitions of redundancy between association rules and provide new characterizations and relationships among them. We show that the main alternatives we discuss correspond actually to just two variants, which differ in the treatment of full-confidence implications. For each of these two notions of redundancy, we provide a sound and complete deduction calculus, and we show how to construct complete bases (that is, axiomatizations) of absolutely minimum size in terms of the number of rules. We explore finally an approach to redundancy with respect to several association rules, and fully characterize its simplest case of two partial premises.

💡 Research Summary

Association rule mining is a cornerstone of modern data‑mining practice, yet the sheer volume of rules generated often obscures the truly informative patterns. This paper tackles the problem from a logical perspective, treating each transaction as a propositional model and defining redundancy as logical entailment: a rule R₂ is redundant if every dataset that satisfies rule R₁ must also satisfy R₂. By re‑examining a variety of existing redundancy notions, the authors demonstrate that, despite superficial differences, they collapse into only two fundamental variants. The distinction lies in how full‑confidence implications are handled.

The first variant, full‑confidence entailment, treats a rule with confidence = 1 exactly as a logical implication. The second, partial‑confidence entailment, restricts entailment to rules whose confidence does not exceed a common threshold θ < 1; both premise and conclusion must respect this bound. This bifurcation clarifies the role of perfect rules and yields two parallel families of inference systems.

For each family the authors present a sound and complete deduction calculus. The calculus consists of a small set of inference rules that preserve confidence and support constraints: (i) augmentation/subset reduction of antecedents, (ii) combination of overlapping antecedents, and (iii) propagation of full‑confidence implications. They prove that any rule that is logically entailed under the chosen variant can be derived using these inference steps, guaranteeing completeness, while each step respects the original statistical thresholds, guaranteeing soundness.

Having a calculus, the next natural question is how to obtain a minimum‑size basis—a smallest possible set of rules from which all others can be derived. The paper shows that the answer hinges on closed itemsets and maximal frequent itemsets. For the full‑confidence case, every closed itemset yields a single implication to its closure; the collection of these implications forms a basis whose size equals the number of closed itemsets, which is provably minimal. For the partial‑confidence case, the authors introduce a hierarchy based on the confidence threshold θ. Within each level they select a minimal set of “generating” rules that dominate all others in that level, again using closed‑set properties to prune redundancies. An algorithm is given that constructs these bases in polynomial time, and empirical tests confirm dramatic reductions in rule count on benchmark datasets.

The final contribution addresses redundancy with multiple premises. While most prior work considers redundancy between a single premise and a single conclusion, the authors explore the simplest non‑trivial scenario: two partial‑confidence premises implying a third rule. They provide a full characterization of when such a multi‑premise entailment holds and extend the deduction calculus to handle it. The key insight is that the intersection of the two antecedents can serve as a new antecedent that, together with the confidence bound, guarantees the conclusion. This result opens the door to more complex multi‑premise reasoning in future work.

In summary, the paper delivers a unified logical theory of association‑rule redundancy, identifies two essential redundancy notions, supplies complete inference systems for each, and constructs absolutely minimal bases in terms of rule count. By bridging statistical thresholds with propositional entailment, it offers both a deeper theoretical understanding and practical tools for producing concise, non‑redundant rule sets, thereby enhancing interpretability and efficiency in data‑mining pipelines.