Study And Implementation Of FCA-based Association Rule Mining

As one of the most active branch of Data Mining, association rule mining aims todiscover the potential, new and human-understand relationships between data items.Concept lattice theory is used to discovery, sort and display the concepts, whose datastructure is concept lattice. Generalization and specialization between concepts in latticecan express knowledge. As a model for representing knowledge, concept lattice is ableto provide strong support for association rule mining.In this paper, we analysis the existing problems in association rule mining. For theproblems like scanning long and too many candidate sets, we propose a new algorithmbased on classical concept lattice, by which all frequent item sets can be obtained by thesearch from top to bottom in lattice. To solve the problem of much redundancy rulesand can’t update dynamically, we optimize the structure of quantitative extended lattice,propose the updates of quantitative extended lattice with the changes in transaction sets,based on which we construct the quantitative extended lattice incrementally; define thepattern of non-redundant rule based on minimum equal intent and closed set and provethe rules formed by the pattern is complete; based on all mentioned above, we presentthe process of incremental non-redundant association rule mining on the quantitativeextended lattice, at the same time we propose the method of mining restricted rules.Through the analysis of execution time in various formal contexts, we verify thatthe algorithm "association rule mining based on the classical lattice" improve theefficiency of Apriori in "search frequent item sets" and "generate association rules andthe module "generate association rules" of "association rule mining based onquantitative extended concept lattice" is better than the other algorithms both inefficiency and quality. By summarizing characteristics of the three algorithms,weobtain that when there is a large number of frequent item sets, the algorithm"association rule mining based on quantitative extended concept lattice" perform bestand when formal context is large and there is small number of frequent item sets, theperformance of "association rule mining based on quantitative extended concept lattice"is worse than that of the other algorithms due to long-time cost in constructing lattice,but it can update with the changes in transaction sets and mine constrained associationrules quickly, so the algorithm is more practical.