Hiding sensitive associative classification rule by data reduction

Abstract

When data sharing becomes necessary, there is a dilemma in preserving privacy. On one hand sensitive patterns such as classification rules should be hidden from being discovered. On the other hand, hiding the sensitive patterns may affect the data quality. In this paper, we present our studies on the sensitive classification rule hiding problem by data reduction approach, i.e., removing the whole selected records. In our work, we focus on a particular type of classification rule, called canonical associative classification rule. And, the impact on data quality is evaluated in terms of the number of affected non-sensitive rules. We present the observations on the data quality based on a geometric model. According to the observations, we can show the impact precisely without any re-computing. This helps to improve the hiding algorithms from both effectiveness and efficiency perspective. Additionally, we present the algorithmic steps to demonstrate the removal of the records so that the impact on the data quality is potentially minimal. Finally, we conclude our work and outline future work directions for this problem. © Springer-Verlag Berlin Heidelberg 2007.