Fuzzy Association Rules Mining And Application

In data mining, association rules mining is one of the important tasks and applied widely. Moreand more researchers focus on it. Up to now, a lot of association rules mining algorithms have beenproposed, amongst them, some classical ones only handle Boolean attributes. However, in reality,there are a large number of quantitative attributes in databases. In order to find the relationshipbetween the quantitative attributes, fuzzy association rules have been introduced. FP-growth is one ofthe algorithms that mine these rules. But, some defects of FP-growth algorithm remain. This thesisfocuses on these defects and optimizes the membership functions. We extend the fuzzy associationrules and propose a generalized fuzzy association rules mining algorithm. The main contents of thisthesis are as follows:（1） The original membership functions limit the application of the fuzzy region division results. Byusing a new fitness function of the genetic algorithm, we optimize the membership functions. Theactual data modifies these membership functions and finally helps us to get better ones. Theoptimization algorithm is applied to the airport noise attribute data set. By using the optimizedmembership functions to fuzz quantitative data, the division of fuzzy region is more reasonable andmore suitable for practical problems. By using the fuzzy attributes database as the input of associationrules mining algorithm, it enhances the credibility of the fuzzy association rules.（2） As we know, when generating fuzzy attribute database, FP-growth algorithm filters out anumber of fuzzy attributes, which loses some useful information. This thesis improves FP-growthalgorithm and proposes a new algorithm which generates more relationship amongst the fuzzyattributes than the original one. The improved algorithm is applied in the classification system.（3） By adding premise attribute weight and the confidence of the conclusion attribute, this thesisproposes generalized fuzzy association rules and their mining algorithm. The premise attribute weightmore clearly reflects the relative importance of the attribute for the conclusion. The confidence of theconclusion attribute reveals the relative credibility of the conclusion. The generalized fuzzyassociation rules have more interpretability than the traditional fuzzy association rules.