Associative classification rules hiding for privacy preservation

摘要

Sensitive patterns could be discovered from the given data when the data are shared between business partners. Such patterns should not be disclosed to the other parties. However, the shared data should be credible and trustworthy for their 'quality'. In this paper, we address a problem of sensitive classification rule hiding by a data reduction approach. We focus on an important type of classification rules, i.e., associative classification rule. In our context, the impact on data quality generated by data reduction processes is represented by the number of false-dropped rules and ghost rules. To address the problem, we propose a few observations on the reduction approach. Subsequently, we propose a greedy algorithm for the problem based on the observations. Also, we apply two-bitmap indexes to improve the efficiency of the proposed algorithm. Experiment results are presented to show the effectiveness and the efficiency of the proposed algorithm.