Mining Top-K High Utility Itemset Using Bio-Inspired Algorithms

Abstract

High utility itemset (HUI) mining is a necessary research problem in the field of knowledge discovery and data mining. Many algorithms for Top-K HUI mining have been proposed. However, the principal issue with these algorithms is that they need to store potential top-k patterns in the memory anytime, and they request the minimum utility threshold to automatically rise when finding HUIs. Consequently, the performance of existing exact algorithms for Top-K HUIs mining tends to decrease when the database size and the number of distinct items in the databases rise. To address this issue, we suggest a Binary Particle Swarm Optimization (BPSO) based algorithm for mining Top-K HUIs effectively, namely TKO-BPSO (Top-K high utility itemset mining in One phase based on Binary Particle Swarm Optimization). The main idea of TKO-BPSO is not only to use a one-phase model and strategy Raising the threshold by the Utility of Candidates (RUC) to effectively increase the border thresholds for pruning the search space but also to adopt the sigmoid function in the updating process of the particles. This might significantly reduce the combinational problem in traditional HUIM when the database size and the number of distinct items in the databases rise. Consequently, its performance outperforms existing exact algorithms for mining Top-K HUIs because it efficiently overcomes the problem of the vast amount candidates. Substantial experiments conducted on publicly available several real and synthetic datasets show that the proposed algorithm has better results than existing state-of-the-art algorithms in terms of runtime, which can significantly reduce the combinational problem and memory usage.