Mining high utility patterns in incremental databases

Chowdhury Farhan Ahmed, Syed Khairuzzaman Tanbeer, Byeong-Soo Jeong, Young-Koo Lee

Abstract: Frequent pattern mining techniques treat all items in the database equally by taking into consideration only the presence of an item within a transaction. However, the customer may purchase more than one of the same item, and the unit price may vary among items. High utility pattern mining approaches have been proposed to overcome this problem. As a result, it becomes a very important research issue in data mining and knowledge discovery. On the other hand, incremental and interactive data mining provides the ability to use previous data structures and mining results in order to reduce unnecessary calculations when the database is updated, or when the minimum threshold is changed. In this paper, we propose a tree structure, called IIUT (incremental and interactive utility tree), to solve these problems together. It uses a pattern growth mining approach to avoid the level-wise candidate set generation-and-test problem, and it can efficiently capture the incremental data without any restructuring operation. Moreover, IIUT has the "build once mine many" property and therefore it is highly suitable for interactive mining. Experimental results show that our tree structure is very efficient and scalable for incremental and interactive high utility pattern mining.