Implementing private k-means clustering using a LWE-based cryptosystem

Combining data analytics with homomorphic encryption is an interesting topic, which finds several applications in the healthcare domain because it enables clients with low computational and/or storage capacity to outsource the analysis of potentially large datasets to the cloud while protecting sensitive data from unwanted access. In this work, we propose a framework for evaluating k-means clustering using a cryptosystem based on the Learning With Errors (LWE) problem. We implemented three variants of this framework in the computer algebra system Sagemath and executed many experiments to test the performance for various values of k-means and LWE parameters.