Privacy-preserving in association rule mining using an improved discrete binary artificial bee colony

Abstract Association Rule Hiding (ARH) is the process of protecting sensitive knowledge using data transformation. Although there are some evolutionary-based ARH algorithms, they mostly focus on the itemset hiding instead of the rule hiding. Besides, unstable convergence to the global optimum solution and designing long solutions make them inappropriate in reducing side effects. They use the basic versions of evolutionary approaches, resulting in inappropriate performance in ARH domain where the search space is large and the algorithms easily get trapped in local optima. To deal with these problems, we propose a new rule hiding algorithm based on a binary Artificial Bee Colony (ABC) approach which has good exploration. However, we improve the binary ABC algorithm to enhance its poor exploitation by designing a new neighborhood generation mechanism to balance between exploration and exploitation. We called this algorithm Improved Binary ABC (IBABC). IBABC approach is coupled with our proposed rule hiding algorithm, called ABC4ARH, to select sensitive transactions for modification. To choose victim items, ABC4ARH formulates a heuristic. The performance of ABC4ARH algorithm on the side effects is demonstrated using extensive experiments conducted on five real datasets. Furthermore, the effectiveness of IBABC is verified using the uncapacitated facility location problem and 0-1 knapsack problem.