Strength of Composition Attacks in Multiple Independent Data Publication

Each organization either government or other collects person-specific data and releases them with a guarantee that sensitive information of an individual remains to conceal. The situation arises when one patient goes two hospitals and the two hospitals publish their collected data independently, an adversary who knows about an individual's visits may infer his/her private information by composition attack. Therefore, privacy preservation for multiple independent data publishing has concerned considerable research interest nowadays. Following this line, the strength of composition attacks requires an in-depth analysis. To meet this gap, in this work we analyze the strength of composition attack concerning the size of the dataset as well as the overlapping ratio of individuals. In addition to these, different values of the anonymization parameter (k of k-anonymization model) have also been analyzed. Our results show that the success of the composition attack has a significant relation with the size of the data set and individual overlapping ratio which assists to seek a new approach to mitigating composition attack in multiple independent data releases.