High Privacy for Data Disclosures Using Tree-EMD

摘要

Now a day's micro data publishing is very useful to the all the organizations that enables the researchers and policy-makers to analyze the data and learn important information. Privacy is a one of the most important factor here. One of the existing methods for privacy measures such as k-anonymity protects against identity disclosures, but it is not providing affective protection against attribute disclosures. Another privacy measure is l-diversity attempts to solve this problem. However, it's neither enough nor economical to forestall attribute disclosures and fails at information utilization. Therefore a base model known as t-closeness and a lot of versatile privacy model known as (n, t)-closeness was developed to archives a decent balance between privacy and utility. The bottom model t-closeness, which needs that the distribution of a sensitive attribute in any equivalence class is near the distribution of the attribute within the overall table (i.e., the space between the 2 distributions ought to be no quite a threshold t). (n, t)-closeness offers higher utility. These closeness measures need likelihood distributions that are assessed victimization Earth Mover's Distance (EMD) measure. We have a tendency to propose to use associate degree economical tree-based rule, Tree-EMD. Tree-EMD exploits the very fact that a basic possible resolution of the simplex algorithm-based problem solver forms a spanning tree. The quantity of unknown variables is reduced to O(N) from O(N2) of the initial EMD. During this paper, we have a tendency to introduce techniques that are implementation of the Tree-EMD and perform advanced experiments to demonstrate its potency.