A Design for Private Data Protection Combining with Data Perturbation and Data Reconstruction

Abstract

With the rapid development of various hardware equipment and saving technology, multiple data with different types are uploaded to saving space. There are some private data can not be ignored. For provider, in order to use and deliver these private data to the third party, data anonymization, such as K-anonymity [1] should be applied to cover the explicit information. For receiver, there are still some way to transform these “fake” data to a new data set which obtain the same statistical properties with the original one while not exactly the same in detailed records. Under this condition, we want to show our work —— data perturbation and data reconstruction to deal with this problem. We use RGADP (Retrievable General Addictive Data Perturbation) [2] to produce data perturbation and EM algorithm to reconstruct data. And the results are Perturbated data can be produced by original data, and it can be delivered, reversed or further reconstructed easily. The reconstructed data still keeps the statistical properties as the original one. Compared with conditional way, this method can be more flexible to adjust the privacy protection degree according to the length of bias interval. We integrated these two process and report on the findings of our experimental evaluation.