An Experimental Study of Distortion-Based Techniques for Association Rule Hiding
Emmanuel D. Pontikakis, Achilleas A. Tsitsonis, Vassilios S. Verykios · Kluwer Academic Publishers eBooks · 2006
Data mining provides the opportunity to extract useful information from large databases. Various techniques have been proposed in this context in order to extract this information in the most efficient way. However efficiency is not our only concern in this study. The security and privacy issues over the extracted knowledge must be seriously considered as well. By taking this into consideration, we study the discovery of association rules in binary data sets and we propose algorithms for selectively hiding sensitive association rules. Association rule hiding is a well researched area in privacy preserving data mining and many algorithms have been proposed to address it. The algorithms that we introduce use a distortion-based technique for hiding the sensitive rules. The hiding process may introduce a number of side effects either by generating rules which were not previously existing (ghost rules) or by eliminating existing non-sensitive rules (lost rules). The proposed algorithms use effective data structures for the representation of the association rules and they strongly rely on the prioritization of the selection of the transactions to choose for falsification (victim transactions) by using weights. In this paper we show that our algorithms perform better than other similar algorithms in this field in eliminating non-sensitive rules without increasing the processing time significantly. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.