Mining top-k high utility itemsets with effective threshold raising strategies

Abstract

Top-K High Utility Itemset (HUI) mining problem offers greater flexibility to a decision maker in specifying her/his notion of item utility and the desired number of patterns. It obviates the need for a decision maker to determine an appropriate minimum utility threshold value using a trial-and-error process. The top-k HUI mining problem, however, is more challenging and requires use of effective threshold raising strategies. Several threshold raising strategies have been proposed in the literature to improve the overall efficiency of mining top-k HUIs. This paper advances the state-of-the-art and presents a new Top-K HUI method (THUI). A novel Leaf Itemset Utility (LIU) structure and a threshold raising strategy is proposed to significantly improve the efficiency of mining top-k HUIs. A new utility lower bound estimation method is also introduced to quickly raise the minimum utility threshold value. The proposed THUI method is experimentally evaluated on several benchmark datasets and compared against two state-of-the-art methods. Our experimental results reveal that the proposed THUI method offers one to three orders of magnitude runtime performance improvement over other related methods in the literature, especially on large, dense and long average transaction length datasets. In addition, the memory requirements of the proposed method are found to be lower.