An Approach to Overcome Inference Channels on k-anonymous Data

The concept of k-anonymity protection model has been proposed as an effective way to protect the identities of subjects in a disclosed database. However, from a k-anonymous dataset it may be possible to directly infer private data. This direct disclosure is called attribute linkage. k-anonymity also suffer to another form of attack based on data mining results. In fact, data mining models and patterns pose a privacy threat even if the k-anonymity is satisfied. In this paper, we discuss how the privacy requirements characterized by kanonymity can be violated by data mining results and introduce an approach to limit privacy breaches. We experiment it by using the adult dataset from the UCI KDD archive. We report the experimental results which show its effectiveness.