A transversal hypergraph approach for the frequent itemset hiding problem

We propose a methodology for hiding all sensitive frequent itemsets in a transaction database. Our methodology relies on a novel technique that enumerates the minimal transversals of a hypergraph in order to induce the ideal border between frequent and sensitive itemsets. The ideal border is then utilized to formulate an integer linear program (ILP) that answers whether a feasible sanitized database that attains the ideal border, exists. The solution of the program identifies the set of transactions that need to be modified (sanitized) so that the hiding can be achieved with the maximum accuracy. If no solution exists, we modify the ILP by relaxing the constraints needed to be satisfied so that the sanitized database preserves the privacy with guarantee but with minimum effect in data quality. Experimental evaluation of the proposed approach on a number of real datasets has shown that the produced sanitized databases exhibit higher accuracy when compared with the solutions of other well-known approaches.