High Utility Itemset Mining by Using Binary PSO Algorithm

Abstract

The goal of pattern mining is to find some novel patterns from a given database. High utility itemset mining (HUIM) is a research direction of the pattern mining as a sub-domain of data mining. It is different from frequent itemset mining (FIM), which does not simply consider the number of occurrences, but considers both the quantity and profit factors of the commodity to reveal profitable products. There have been some algorithms that are used to mine high utility itemsets (HUIs). This thesis proposes a binary particle swarm algorithm (BPSO) with V-shaped transfer function and nonlinear acceleration coefficient strategy. The original BPSO algorithm lacks local search capabilities in the later stage, which will result in not enough HUIs to be mined. The transfer function in the BPSO algorithm determines the probability of the position change of a particle during the iteration. The transfer function used in the original BPSO algorithm is sigmoid function, which does not sufficiently reflect the probability between the velocity and position change of the particles, but the V-shaped transfer function solves this problem. And considering the influence of acceleration factor on particle motion mode and trajectory, a nonlinear acceleration strategy is used to enhance the search ability of particles. And the dimensions of the item are reduced before the mining process to reduce the redundant combinations in the iterative process. Experiments show that the proposed algorithm outperforms the original BPSO algorithm in terms of the number of mined HUIs.