Efficient Multidimensional Suppression for K-Anonymity

Abstract

Abstract-Many applications that employ data mining techniques involve mining data that include private and sensitive information about the subjects. One way to enable effective data mining while preserving privacy is to anonymize the dataset that include private information about subjects before being released for data mining. One way to anonymize dataet is to manipulate its content so that the records adhere to k-anonymity. Two common manipulation techniques used to achieve kanonymity of a dataset are generalization and suppression. Generalization refers to replacing a value with a less specific but semantically consistent value, while suppression refers to not releasing a value at all. Generalization is more commonly applied in this domain since suppression may dramatically reduce the quality of the data mining results if not properly used. However, generalization presents a major drawback as it requires a manually generated domain hierarchy taxonomy for every quasiidentifier in the dataset on which k-anonymity has to be performed. In this paper we propose a new method for achieving kanonymity named K-anonymity of Classification Trees Using Suppression (kACTUS). In kACTUS efficient multi-dimensional suppression is performed, i.e., values are suppressed only on certain records depending on other attribute values, without the need for manually-produced domain hierarchy trees. Thus, in kACTUS we identify attributes that have less influence on the classification of the data records and we suppress them if needed in order to comly with k-anonymity. The kACTUS method was evaluated on ten separate datasets to evaluate its accuracy as compared to other k-anonymity generalization and suppressionbased methods. Encouraging results suggest that kACTUS' predictive performance is better than that of existing k-anonymity algorithms. Specifically, on average the accuracies of TDS, TDR and kADET are lower than kACTUS in 3.5%, 3.3% and 1.9% respectively despite their usage of manually defined domain trees. The accuracy gap is increased to 5.3%, 4.3% and 3.1% respectively when no domain trees are used.

kACTUS 2: Privacy Preserving in Classification Tasks Using k-Anonymity

“…The proposed method requires no prior knowledge and can be used by any inducer. It combines compensation by suppression presented in [15] and compensation by swapping which decreases information loss induced by the suppression approach. The new method also shows a higher predictive performance and less information loss when compared to existing stateof-the-art methods.…”

“…In our previous research [15], we presented a k-anonymity classification treebased suppression (kACTUS). kACTUS wraps a decision tree inducer which is used to induce a classification tree from the original dataset.…”

Identity Disclosure Protection in Dynamic Networks Using KW – Structural Diversity Anonymity

“…It is K-anonymity of Classification Trees Using Suppression(kACTUS) [5] model which performs multidimensional suppression of certain records which depends on other attribute values, without any need of domain hierarchy trees that are produced manually.…”

A Survey of Privacy Preserving Data Publishing using Generalization and Suppression

“…It is without any doubts that there exists some theory or techniques to go on data anonymization only using suppression operation [37,38,39]. Like the category of generalization, there are five kinds of suppression, attribute suppression [34], record suppression [40], value suppression [41], cell suppression [42] and multidimensional suppression [43]. Attribute suppression suppresses the whole values of the attribute.…”

“…For this reason, general metric leverages all kinds of factors, make it fit for different situation as much as possible by measuring discrepancy of anonymous and original dataset. In early works [42,43,44,45], the definition of data metric is not totally formed, and it just has the concept of minimal distortion that corresponding algorithm must comply with. For example, one measure of information loss was the number of generalized entries in the anonymous dataset and some algorithm judges the data quality by the ratio of the size of anonymous and original dataset.…”