Distortion-based privacy-preserved association rules mining without side effects using closed itemsets

Abstract

Privacy-Preserving Data Mining (PPDM) is taking more focus in the recent years due to growing concern about privacy. Association rule mining is a widely used data mining technique. The goal of privacy-preserved association rule mining is to protect sensitive rules from disclosing in the published data. It is achieved by reducing the confidence below a minimum threshold or adding noise to the original confidence of the sensitive rules. Existing data sanitization techniques suffer from side effects which reduce the utility of the data. We propose a model-based approach for preserving the privacy of sensitive rules without side effects. The proposed value distortion-based sanitization algorithm sanitizes the closed itemsets instead of transactions of the database. Also, the proposed solution is scalable for distorting the whole database without changing the relationship between attribute values or itemsets in the original database. Experimental results show that the proposed method is better than other well-known techniques based on transaction modification concerning the sanitization time and side effects.