Privacy Preserving Frequent Itemsets Mining Based on Database Reconstruction

Privacy preserving frequent itemsets mining (PP-FIM) aims at transforming a database so as to efficiently achieve frequent itemsets mining without revealing any sensitive knowledge. However, the majority of the proposed PPFIM methods are based on the idea of sanitizing database. The conflict between knowledge mining and privacy preserving is hard to avoid. To this end, we propose a novel PPFIM algorithm based on database reconstruction called DR-PPFIM, which can afford high data utility as well as high degree of privacy. In DR-PPFIM, a sanitization algorithm is first performed to remove all sensitive knowledge. Then a novel database reconstruction scheme is designed to reconstruct a new database based on the remained non-sensitive frequent itemsets. In addition, we propose a further hiding strategy to further decrease the importance of sensitive itemsets so that the threat of disclosing confidential knowledge can be reduced. Experimental evaluations of the proposed DR-PPFIM on real datasets are reported to show the superiority of DR-PPFIM compared with other state-of-the-art algorithms.