Privacy preserving data mining with 3-D rotation transformation

Abstract

Data perturbation is one of the popular data mining techniques for privacy preserving. A major issue in data perturbation is that how to balance the two conflicting factors – protection of privacy and data utility. This paper proposes a Geometric Data Perturbation (GDP) method using data partitioning and three dimensional rotations. In this method, attributes are divided into groups of three and each group of attributes is rotated about different pair of axes. The rotation angle is selected such that the variance based privacy metric is high which makes the original data reconstruction difficult. As many data mining algorithms like classification and clustering are invariant to geometric perturbation, the data utility is preserved in the proposed method. The experimental evaluation shows that the proposed method provides good privacy preservation results and data utility compared to the state of the art techniques. Keywords: Data perturbation, Variance, Three dimensional rotation, Privacy preserving, Data mining