| Summary: | 碩士 === 國立高雄大學 === 資訊工程學系碩士班 === 100 === Data mining techniques have been widely applied to various business and research issues. Since traditional quantitative rule mining only considers the occurrence and quantity interval relationships of items in transactions, fuzzy itemset mining was proposed to consider the quantity of items and make the quantitative rules that are simple and thus more comprehensible to decision makers. However, most existing fuzzy mining techniques adopt level-wise techniques to deal with the problem of fuzzy itemset mining, and thus the performance of the existing algorithms is not very good. To address this, in this thesis, we thus develop two efficient methods, GDF (Gradual Data-Reduction Fuzzy Mining Approach) and PFA (Projection-based Fuzzy Mining Approach), to speed up the execution efficiency of finding fuzzy frequent itemsets. In particular, the two approaches proposed, GDF and the PFA, adopt a data-reduction strategy consisting of pruning and merging processes, as well as two other strategies, indexing and filtering, to effectively reduce the number of unpromising candidate itemsets in comparison with the existing algorithms. The results of an experimental evaluation reveal that the proposed approaches can achieve up to a 50% improvement in efficiency over the traditional fuzzy mining algorithm on several datasets.
|
|---|
