Privacy preserving association rule mining using attribute-identity mapping

Abstract

Association rule mining uncovers hidden yet important patterns in data. Discovery of the patterns helps data owners to make right decision to enhance efficiency, increase profit and reduce loss. However, there is privacy concern especially when the data owner is not the miner or when many parties are involved. This research studied privacy preserving association rule mining (PPARM) of horizontally partitioned and outsourced data. Existing research works in the area concentrated mainly on the privacy issue and paid very little attention to data quality issue. Meanwhile, the more the data quality, the more accurate and reliable will the association rules be. Consequently, this research proposed Attribute-Identity Mapping (AIM) as a PPARM technique to address the data quality issue. Given a dataset, AIM identifies set of attributes, attribute values for each attribute. It then assigns ‘unique’ identity for each of the attributes and their corresponding values. It then generates sanitized dataset by replacing each attribute and its values with their corresponding identities. For privacy preservation purpose, the sanitization process will be carried out by data owners. They then send the sanitized data, which is made up of only identities, to data miner. When any or all the data owners need(s) ARM result from the aggregate data, they send query to the data miner. The query constitutes attributes (in form of identities), minSup and minConf thresholds and then number of rules they are want. Results obtained show that the PPARM technique maintains 100% data quality without compromising privacy, using Census Income dataset.