Mining association rules from distorted data for privacy preservation

Abstract

In order to improve privacy preservation and accuracy, we present a new association rule mining scheme based on data distortion. It consists of two steps: First, the original data are distorted by a new randomization method. Then, the mining algorithm is implemented to find frequent itemsets from the distorted data, and generate association rules. With reasonable selection for the random parameters, our scheme can simultaneously provide a higher privacy preserving level to the users and retain a higher accuracy in the mining results. © Springer-Verlag Berlin Heidelberg 2005.