General-purpose multi-user privacy-preserving outsourced k-means clustering

Abstract

Nowadays, there is a trend towards an incessant growth and complexity in the volume of data held by users. Clustering techniques in machine learning are becoming more and more important to help extract the value from big data. However, it is difficult for a single user to fully use large-scale data for clustering locally due to the restricted training resources and the lack of datasets. To address this problem, the multi-user collaborative clustering model has emerged as a viable solution for multi-user collaborative clustering by hosting the data on a cloud platform. Nevertheless, outsourced clustering may give rise to a series of privacy problems. In order to address these problems effectively, we propose a novel, secure and efficient outsourced k-means clustering scheme. This scheme uses partially homomorphic encryption techniques for cloud-based k-means clustering, which ensures that the cloud does not contain any private information, while simultaneously safeguarding the confidentiality of the database, data involved in the clustering process, clustering results and user information. Furthermore, a comparative analysis of our proposed scheme is conducted. These analyses demonstrate the security and practicality of our scheme.