Optimizing association rule hiding using combination of border and heuristic approaches

Data sanitization process transforms the original database into a modified database to protect the disclosure of sensitive knowledge by reducing the confidence/support of patterns. This process produces side-effects on the sanitized database, where some non-sensitive patterns are lost or new patterns are produced. Recently, a number of approaches have been proposed to minimize these side-effects by selecting appropriate transactions/items for sanitization. The heuristic approach is applied to hide sensitive patterns both in association rules and in frequent itemsets. On the other hand, the border, exact, and evolutionary approaches have only been designed to hide frequent itemsets. In this paper, a new hybrid algorithm, called Decrease the Confidence of Rule (DCR), proposed to improve a border-based solution, namely MaxMin, using two heuristics to hide the association rules. To achieve this, first, a heuristic was formulated in combination with MaxMin solution to select victim items in order to control the impact of sanitization process on result quality. Then, the victim items were removed from transactions with the shortest length. Some experiments have been conducted on the four real datasets to compare performance of DCR with the Association Rule Hiding based on Intersection Lattice (ARHIL) algorithm. The experimental results showed that the proposed algorithm yielded fewer side-effects than ARHIL algorithm. In addition, its efficiency was better than the heuristic approach.