EA-HUFIM: Optimization for Fuzzy-Based High-Utility Itemsets Mining

Abstract

High-utility itemsets mining (HUIM) has attracted considerable attention due to its superior performance in discovering the profitable itemsets from databases. Generally, an itemset is discovered as a high-utility itemset (HUI) by HUIM only if its utility (such as profit) is higher than a predefined utility threshold. The previous HUIM methods cannot address the sharp boundary problem and the efficiency problem simultaneously. To tackle the above problems, we take the attempt to combine the fuzzy set theory and the evolutionary algorithms (EAs) in the HUIM method, which is called EA-HUFIM. For the proposed EA-HUFIM, it is challenging to design an appropriate EA and a suitable fitness function. Therefore, we focus on analyzing the characteristics of fuzzy-based HUI (i.e., HUFI) and present a theorem and a lemma based on a new concept, the high database-max utility fuzzy itemsets (HDFIs). The fitness function adopted in the EA-HUFIM method is inspired by the proposed lemma. Besides, we present a general EA framework with considering the elite learning strategy. The top K 1-HUFIs and top K 1-HDFIs are stored as an elite archive for further offspring production. To improve efficiency, we further employ a fuzzy map and a transaction map to propose a novel preprocessing method, and thus, the computational costs can be reduced. Experiments on several datasets and the results demonstrate that the EA-HUFIM method is effective in quantitative databases.