A Privacy-Preserving Data Obfuscation Scheme Used in Data Statistics and Data Mining

Many applications are benefited from data sharing, especially data statistics and data mining. But as the shared data may contain private information of data owner, it has a high risk of revealing data owner\´s privacy. Data obfuscation is proposed to gain a balance between data privacy and data usability. But it is hard for the present obfuscation schemes to remain the usability of data in a fine-grained level. Besides, the original data can\´t be retrieved from the obfuscated data. To address the above issues, we proposed a data obfuscation scheme that adds an accurate "noise" to the original data to protect the privacy while keeping the numeral characteristics of data unchanged in different levels. Besides, the scheme can also lower the impact on data mining. Furthermore, by allocating different keys to users, different users have different permissions to access to data. To achieve this, our scheme comes in four steps. Firstly, an improved cloud model is proposed to generate an accurate "noise". Next, an obfuscation algorithm is propose to add noise to the original data. Then, an initial scheme for dataset obfuscation is proposed, including the grouping and key allocating processes. In the final step, a fine-grained grouping scheme based on similarity is proposed. The experiments show that our scheme obfuscates date correctly, efficiently, and securely.