More Efficient Algorithms for Mining High-Utility Itemsets with Multiple Minimum Utility Thresholds

Abstract

Mining high-utility itemsets (HUIs) is a popular data mining task, which consists of discovering sets of items that yield a high profit in a transaction database. Although HUI mining has numerous applications, a key limitation is that a single minimum utility threshold (minutil) is used to assess the utility of all items. This simplifying assumption is unrealistic since in real-life all items do not have the same unit profit, and thus do not have an equal chance of generating a high profit. As a result, if the minutil threshold is set high, patterns containing items having a low unit profit are often missed, while if minutil is set low, the number of patterns becomes unmanageable. To address this issue, this paper presents an efficient tree-based algorithm named HIMU for mining HUIs using multiple minimum utility thresholds. A novel tree structure called multiple item utility Set-enumeration (MIU)-tree and the global and conditional downward closure (GDC and CDC) properties of HUIs in the MIU-tree are proposed. Moreover, a vertical compact utility-list structure is adopted to store the information required for discovering HUIs without performing additional database scans and generating candidates. An extensive experimental study on real-world and synthetic datasets show that this greatly improves the efficiency of the algorithm in terms of runtime and scalability.