Article Preview
Top1. Introduction
Data mining field that aims to find interesting patterns from huge amounts of data attracted the increasing interest of organizations, researchers and practitioners. However this growing use of data mining technology in different domains increased the concern for privacy leading to an active research area named privacy preserving data mining. From a general point of view, privacy issues related to the application of data mining can be classified into two main categories, namely data hiding and knowledge hiding. Data hiding aims to remove confidential or private information from data prior to its publication. Knowledge hiding, on the other hand is concerned with the sanitization of data leading to disclosure of confidential and private knowledge (Gkoulalas-Divanis and Verykios, 2010). The problem of knowledge hiding requires sanitizing the input database D in such a way that a set of sensitive knowledge is hidden, while most of the information in D is maintained.
Association rule mining or frequent itemset mining is a major data mining methodology, mostly known in the area of market basket analysis and aims to capture relationships present among items in a transactional database (Agrawal et al., 1993) (Agrawal and Srikant, 1994). Despite its benefit in modern business, frequent itemset mining can also pose a threat to privacy and security in a database sharing environment when precautions are not taken in its implementation (Atallah et al., 1999) (Oliveira and Zaiane, 2002). Frequent itemset hiding is specialization of the generic knowledge hiding problem where the main requirement asks for lowering the support of sensitive itemsets in the input database D so that sanitized database can be produced. Secondary requirements are minimization of deleted items and loss of non-sensitive frequent itemsets (Bonchi and Ferrari, 2011).
Large body of research emerged in the field of itemset hiding, since the database owners are in need of sharing data with their competitors, for their mutual benefit without revealing strategic patterns in the form of sensitive itemsets. Due to combinatorial nature of the problem of itemset hiding, proposed sanitization methodologies span from simple, time and memory efficient heuristics (Oliveira and Zaiane, 2002) (Oliveira and Zaiane, 2003a) (Verykios et al., 2004) (Amiri, 2007) (Wu et al., 2007) (Keer and Singh, 2012) (Yildiz and Ergenc, 2012), border-based approaches (Sun and Yu, 2005) (Sun and Yu, 2007) (Moustakides and Verykios, 2008) and reconstruction based approaches (Mielikainen, 2003) (Guo, 2007) (Lin and Liu, 2007) (Boora et al., 2009) (Mohaisen et al., 2010) to exact hiding (Menon et al., 2005) (Gkoulalas-Divanis and Verykios, 2006) (Gkoulalas-Divanis and Verykios, 2008) (Gkoulalas-Divanis and Verykios, 2009b) algorithms that offer guarantees on the quality of the computed hiding solution at an increased computational complexity cost. Whatever the technique used in sanitization; different attributes are used in selecting the transaction, itemset in the transaction or the item in the itemset to modify. Sanitization techniques try to minimize distance and/or information loss while the number of sensitive itemsets, characteristics of the data sets or user defined support vary.