Association Rules Hiding via Multi-Objective Differential Evolution Algorithm

摘要

Association rules (ARs) mining has been widely used in discovering interesting patterns from massive data as a key data mining technique. However, this mining technique may lead to the disclosure of sensitive ARs, and thus, the ARs hiding strategy is very important and necessary. In this paper, a novel ARs hiding method (DEvoArH) is proposed for hiding the ARs. Noting that there are three side-effects in ARs hiding (ARH), i.e., the hiding failure rate, the lost rules rate, and the ghost rules rate, a multi-objective differential evolution algorithm is carefully designed. Specifically, the Pareto-optimal solutions are defined as the elite set, and then each solution is capable of learning from the elite set with a certain probability. Considering the convergence speed and search ability, a directional mutation and random mutate round is introduced into the mutation operator. Besides, to further decrease the data change rate, a database pre-processing mechanism is proposed to filter the unrelated data before the hiding process. Finally, the results of the comparison experiments also demonstrate the superiority of the proposed ARH method in terms of the three side-effects and efficiency.