New approaches for mining high utility itemsets with multiple utility thresholds

Abstract

Recently, two research directions have been noticed in data mining: frequent itemset mining (FIM) and high utility itemset mining (HUIM). The FIM process will output itemsets whose number of occurrences together exceeds or equals the required threshold, but this process ignores the beneficial attribute of each item. HUIM algorithms are proposed to overcome the disadvantage of FIM, but these algorithms only use a single threshold, which is unsuitable in the real world when applications often require different utility thresholds. HUIM algorithms with multi-threshold utilities are proposed, but these have high mining time and memory consumption. This paper thus presents an efficient method for Mining High Utility Itemsets with Multiple Utility Thresholds (MHUI-MUT). The article applies upper bounds and the strategy of pruning, thus reducing database scanning, and proposes a cut-off threshold to minimize the mining time.We also present a method to parallelize the algorithm to make the most of the performance of multi-core computers. The experimental results show the superior speed of the MHUI-MUT algorithm compared to the previous one, and the parallel version also outperforms the proposed sequential algorithm.