Minging Of Skyline Patterns By Considering Both Frequent And Utility Constraints

Abstract

Mining frequent itemsets (FIM) are the task to find the itemsets that are frequently occurrence in a customer transaction database. However, FIM ignores the weight, interestingness or unit profit of the items. To unveil more details, the task of High-Utility Itemset Mining is proposed, as a more generalized task than FIM, to reveal the items that has high profit (or utility) from transaction databases. By considering both utility and frequency measure, SKYMINE algorithms used to find Patterns of Skyline Frequency-Utility (named SFUPs). However, SKYMINE has the disadvantage that it takes a lot of computations to find SFUPs. In 2019 Lin et al presented an algorithm called SKYFUP-D used to mine SFUP. However, dense databases, which are the ones containing several similar transactions, have negative impact on the SKYFUP-D’s performance both in runtime and memory usage. Therefore, the thesis proposes an algorithm called MSKY-D, as an extension to the original SFYFUP-D algorithm, to utilize a technique called transaction merging. The proposed approach merges similar transactions in a transaction database to reduce the cost of database scans, candidates checking and memory usage. Experiment evaluations also show that the MSKY-D algorithm has better performance in terms of time and memory than the SKYFUP-D algorithm, especially on dense databases.