Mining High-Utility Irregular Itemsets

Abstract

High-utility itemset mining (HUIM) currently plays an important role in a wide range of applications and data mining community. Several algorithms, methods and data structures have been proposed to improve efficiency of mining for such itemsets. Besides, HUIM is extended in several aspects including the regarding of “regularity or irregularity of occurrence” on high utility itemsets. This leads to the emerging of high-utility regular itemsets mining (HURIM) and high-utility irregular itemsets mining (HUIIM) which can help to observe occurrence behavior of high utility itemsets. Based on HURIM, the regularity threshold is applied to measure interestingness of itemsets and to prune search space. However, on HUIIM, the threshold cannot help to prune uninteresting itemsets causing this task consumes high computational cost. Thus, in this paper, we here present a single-pass algorithm, called HUIIM (High-Utility Irregular Itemset Miner), for efficiently mining high-utility irregular itemsets. The new-modified utility list structure (NUL) is applied for maintaining occurrence information simultaneously with utility value of an itemset and also for fast calculation of total utility of the itemset. Moreover, a new pruning technique is designed and applied to improve computational performance of HUIIM. Experimental studies were conducted on synthetic and real datasets to show efficiency of HUIIM (with and without the new pruning technique) in the terms of computational time and memory usage.