Preserving Privacy of Sensitive Itemsets using Controlled Perturbation of Closed Itemsets

摘要

Data perturbation is one of the famous techniques in privacy preserved data mining. It is considered relatively easy and effective approach for preserving sensitive information in the released data. In this paper, the authors propose an improved version of their previous work which uses value distortion-based data sanitization algorithm to safely perturb the original support of sensitive itemsets without generating any side effects. The data sanitization method used in the previous work suffers from spurts of uncontrolled loss of information and support accuracy of itemsets in sparse datasets. To overcome this limitation, in this paper, the authors propose an improved technique which randomly distorts the support of sensitive itemsets in the closed itemset lattice within specified error limit or threshold, also keeping the relationship between itemsets unchanged. Experimental results show that the proposed improved approach is more efficient in perturbing the data to preserve privacy when compared with the previous work and other well-known distortion-based approaches.