Frequent and High Utility Itemsets Mining Based on Bi-Objective Evolutionary Algorithm with An Improved Mutation Strategy

Abstract

Frequent and high utility itemsets mining (FHUIM) is one of the important tasks in pattern mining. In order to solve the exponential search space and parameter setting problems that traditional HUIM algorithms encountered, the task of FHUIM was reformulated as a bi-objective problem that can be solved by multi-objective evolutionary algorithms (MOEAs). However, the search efficiency of the MOEAs may become lower when the total distinct items, the number of transactions, and the average length of transactions in the database are larger. To further improve the efficiency of MOEAs for FHUIM, we proposed FHUIM based on bi-objective evolutionary algorithm with an improved mutation strategy (FHUIM-BOEA-IMS). In FHUIM-BOEA-IMS, an improved mutation strategy is proposed to make the items with higher support and utility more likely to be saved in population, by which the FHUIs are more likely to be searched. The results on four popular datasets show that the proposed FHUIM-BOEA-IMS has better performance than the compared baseline in the task of FHUIM in terms of the convergence and final solutions.