Data anonymization: k-anonymity and de-anonymization attacks
The need for access to data concerning individuals is steadily increasing the last years. These data may be collected by governments and companies for various reasons and purposes. However, the release of data may cause several issues if we don’t take the appropriate actions. There is high risk o...
Saved in:
| Summary: | The need for access to data concerning individuals is steadily increasing the last years. These
data may be collected by governments and companies for various reasons and purposes.
However, the release of data may cause several issues if we don’t take the appropriate actions.
There is high risk of disclosure of confidential information concerning individuals.
Modifications to the data must be applied in order to protect individuals’ privacy, but
attention must be paid in order to keep a good balance between the desired level of privacy
and data utility. Different anonymization methods have been proposed and the selection of
which to apply depends on several variables: the type of release, the data type (e.g. numerical,
nominal, ordinal), and the desired level of disclosure limitation. k-anonymity was the first
privacy model that has been proposed and consists the basis for many other models that
followed. However it is vulnerable to several attacks and needs enforcement. In this thesis
there is a thorough presentation of the majority of anonymization models classified by the
type of disclosure they tried to protect. Most of the models that are presented here are
extensions of k-anonymity model but some different approaches are also presented for better
understanding of its shortcomings. The algorithms that have been proposed so far for kanonymity
implementation have been also presented in a novel classification. Furthermore,
applications of k-anonymity in several fields are examined. An allusion to different types of
attacks to released data is maid, attack algorithms and their metrics are presented and a new
de-anonymization tool with a series of new attack algorithms and their metrics is proposed. |
|---|