Hiding sensitive itemsets with multiple objective optimization

Jerry Chun Wei Lin*, Yuyu Zhang, Binbin Zhang, Philippe Fournier-Viger, Youcef Djenouri

*Corresponding author for this work

Research output: Contribution to journalJournal articleResearchpeer-review

Abstract

Privacy-preserving data mining (PPDM) has become an important research topic, as it can hide sensitive information, while ensuring that information can still be extracted for decision making. While performing the sanitization progress for hiding the sensitive information, three side effects such as hiding failure, missing cost, and artificial cost happen at the same time. Several evolutionary algorithms were introduced to minimize those three side effects of PPDM using a single-objective function that generates one solution for sanitization. This paper presents a multiobjective algorithm (NSGA2DT) with two strategies for hiding sensitive information with transaction deletion based on the NSGA-II framework. To obtain better balance of side effects, the designed NSGA2DT takes database dissimilarity (Dis) as one more factor to achieve better performance in terms of four side effects. Moreover, instead of a single solution of the sanitization progress, the designed NSGA2DT provides more than one solutions than those of single-objective evolutionary algorithms, which shows flexibility to select the most appropriate transactions for deletion depending on user’s preference. A Fast SoRting strategy (FSR) and the pre-large concept are utilized, respectively, in this paper to find the optimized transactions for deletion and speed up the iterative process. Based on the developed NSGA2DT, the set of several Pareto solutions can be easily discovered, thus avoiding the problem of local optimization of single-objective approaches. Besides, the designed NSGA2DT does not require to set initial weights for evaluating the side effects, and thus, the results could not be seriously influenced by the predefined weights. Experimental results show that the proposed NSGA2DT provides satisfactory results with reduced side effects, compared to previous evolutionary approaches with single-objective function.

Original languageEnglish
JournalSoft Computing
Volume23
Issue number23
Pages (from-to)12779-12797
Number of pages19
ISSN1432-7643
DOIs
Publication statusPublished - 1. Dec 2019

Keywords

  • Evolutionary computation
  • Pareto solutions
  • PPDM
  • Pre-large concept
  • Sanitization

Fingerprint Dive into the research topics of 'Hiding sensitive itemsets with multiple objective optimization'. Together they form a unique fingerprint.

Cite this