Privacy-Preserving Association Rules Mining Using the Grouping Unrelated-Question Model

Privacy preserving data mining algorithms have been intensively investigated since 2000. But almost all these approaches don´t consider different privacy attitudes of different people. Some people are willing to provide their privacy data to investigators under almost any condition. Thus the privacy- preserving algorithm grouping informants will lead to higher accuracy of results. The Warner model, an important algorithm on collecting privacy data, limits the probability of randomized response and two questions in the model are closely related, which decreases informants´ cooperation. Hence we propose the grouping unrelated-question model (GUQM) for the first time to solve the limits of the Warner model and to collect privacy data of people with different attitudes about privacy. Furthermore, we combine association rules mining with GUQM to solve the high time complexity in association rules mining. The paper analyzes the algorithm´s validity, privacy protection and time complexity in theory. The experimental results show the effect of every parameter in our algorithm on the results of mining association rules.