An efficient method for mining high average-utility itemsets based on particle swarm optimization with multiple minimum thresholds

Abstract

In the world of data exploitation, high average-utility itemset mining (HAUIM) is essential. In contrast to traditional high-utility itemset mining, HAUIM does not favor long itemsets. Although most HAUIM techniques are useful for finding high average-utility itemsets (HAUIs) with a unique minimal utility threshold, their applicability to real-world data analysis is limited. For HAUI mining, two MEMU and HAUIM-MMAU algorithms were previously developed using a variety of minimum high-average utility values. Nevertheless, they are inefficient since MEMU uses the average-utility list structure for HAUI mining, while HAUIM-MMAU is based on a solution that generates and tests candidates. The two effective HAUIM algorithms that have been developed to address this issue are HAUI-BPSO-MMAU (Mining High Average-Utility Itemset utilizes Binary Particle Swarm Optimization with Many Minimum Average-Utility values) and HAUIF-PSO-MMAU (Mining High Average-Utility Itemset uses a new bio-HAUI Framework of Particle Swarm Optimization for mining HAUIs with Many Minimum Average-Utility values). Both algorithms are based on particle swarm optimization (PSO) with various minimum average utility values. To explore HAUIs with various minimum average utility values, we present a sorting technique that restores the average-utility upper bound (AUUB) property. We also introduce two effective pruning techniques to enhance the exploitation performance for mining HAUIs and reduce the search space. Extensive tests on five publicly accessible basic datasets indicate that the proposed algorithms outperform MEMU and HAUIM-MMAU algorithms in terms of runtime and memory usage.