IHUMN: an improved high-utility itemsets mining algorithm with negative utility items

Abstract

High-utility itemset mining is to mine high profit itemsets from transaction databases. But if there are some itemsets with negative utility values in the transaction database, the high-utility itemsets with the negative values may be pruned incorrectly and the subset of the low-utility itemsets may be the high-utility itemsets. In this paper, an improved high-utility itemsets mining algorithm with negative utility items (IHUMN) is proposed. A novel utility-list buffer structure with negative unit profits is proposed to efficiently store and retrieve utility-list, and reduce the memory consumption during the mining process. Moreover, Transitive Extension with Negative utility formula is constructed to compute the upper bound of utility avoiding the overestimation of low-utility itemsets as high-utility itemsets. The performance of IHUMN is evaluated, and compared against the FHN and GHUM method. The results of the experiments confirm that IHUMN has a favorable improvement in terms of time costs, the memory utilization and the number of visited nodes. The IHUMN algorithm consumes 40% less memory than GHUM. Moreover, the algorithm has good performance on dense datasets.