Enhancing threshold-raising strategies for effective mining top-k high utility patterns

Mining high utility patterns, the subject of which has attracted many researchers in data mining, is the process of discovering patterns with utility satisfying a minimum predetermined threshold. Many studies have been performed, but finding the suitable minimum utility threshold is problematic, because users cannot predict the appropriate threshold that affects mining performance. To solve this problem, mining the highest utility of k patterns, called top-k high utility, has been proposed. Although many approaches have been proposed, the issue of many candidates and the performance of mining needed to be further studied. In this paper, we propose a top-k high utility mining method that does not produce a candidate with an effective threshold-raising strategy. Instead, the proposed method uses a utility-list data structure with improved threshold-raising strategies combined with an efficient pruning strategy. Experimental results on real and synthesis datasets show that the algorithm presented performs better than current methods.