Encrypted data indexing for the secure outsourcing of spectral clustering

Spectral clustering is one of the most popular clustering methods and is particularly useful for pattern recognition and image analysis. When using spectral clustering for analysis, users are either required to implement their own platforms, which requires strong data analytics and machine learning skills, or allow a third party to access and analyze their data, which may compromise their data privacy or security. Traditionally, this problem is solved by privacy-preserving data mining using randomization perturbation or secure multi-party computation. However, the existing methods suffer from the problems of inaccurate results or high computational requirements on the data owner’s side. To address these problems, in this paper, we propose a new secure outsourcing data mining (SODM) paradigm, which allows data owners to encrypt their data to ensure maximum data security. After the encryption, data owners can outsource their encrypted data to data analytics service providers (i.e., data analytics agent) for knowledge discovery, with a guarantee that neither the data analytics agent nor the other parties can compromise data privacy. To allow data mining to be efficiently carried out on encrypted data, we design a secure KD-tree to index all the encrypted data. Based on the SODM framework, a secure spectral clustering algorithm is proposed. The experiments on real-world datasets demonstrate the effectiveness and the efficiency of the system for the secure outsourcing of data mining.