A DCT-based privacy-preserving approach for efficient data mining

With the rapid growth of various data collected by companies and organizations, there is an increasing need for the data owners to share their data with the third party for the purpose of data mining. Therefore, protecting the privacy of the shared data has received considerable attention from academia and industry. However, existing methods do not work well for distance-based mining algorithms. This paper proposes a novel privacy-preserving approach to support distance-based mining algorithms while reducing the shared data size. This approach transforms the shared data to discrete cosine transformation (DCT) coefficients, and adaptively selects DCT coefficients so as to form the compressed coefficient matrix. To enhance data privacy, this approach employs rotation-based transformation with modified constraints to process the compressed coefficient matrix. Extensive experiments show that the proposed approach achieves the best tradeoff between data privacy and mining quality comparing with other existing ones.