GENERALIZATIONS WITH PROBABILITY DISTRIBUTIONS FOR DATA ANONYMIZATION
Mehmet Ercan Nergiz, Suleyman Cetintas, Ferit Akova · Purdue e-Pubs (Purdue University System) · 2008
Anonymization based privacy protection ensures that data cannot be traced to an individual.Many anonymify algorithms proposed so far made use of d~fferent value generalization techniques to satisfy d~jferent privacy constraints.This paper presents pdf-generalization merhod that empowers data value generalizations with probability distribution functions enabling the publisher to have better control over the trade off between privacy and utilization.We evaluate the pdf approach for k-anonymity and 6-presence privacy models and show how to use pdf generalizations to utilize datasets even further without violating privacy constraints.Paper also shows theoretically and experimentally that information gained from pdfs increases the utilization of the anonymized data w.r.t.real world applications such as class$cation and association rule mining.