K-anonymous association rule hiding

In the paper we point out that the released dataset of an association rule hiding method may have severe privacy problem since they all achieve to minimize the side effects on the original dataset. We show that an attacker can discover the hidden sensitive association rules with high confidence when there is not enough "blindage". We give a detailed analysis of the attack and propose a novel association rule hiding metric, K-anonymous. Based on the K-anonymous metric, we present a framework to hide a group of sensitive association rules while it is guaranteed that the hidden rules are mixed with at least other K-1 rules in the specific region. Several heuristic algorithms are proposed to achieve the hiding process. Experiment results are reported to show the effectiveness and efficiency of the proposed approaches.