Privacy preserving k-means clustering : a survey research

Abstract EN

The clustering is an exploratory task of data mining.

This raised a real problem of privacy when the data are from different sources.

Most of researches on privacy preservation in clustering are developed for k-means clustering algorithm, by applying the secure multi-party computation framework.

The distribution of data may be different (vertical, horizontal or arbitrary).

Approaches allowing solving the problem on a vertical, horizontal and even arbitrary partitioned dataset were proposed.

The major interest is to reveal the minimum of information during the execution of the algorithm, especially in k-means iterations, which poses a real challenge for secure multiparty computation.

This work consists to study and analyze all works of privacy preserving in the k-means algorithm, classify the various approaches according to the used data distribution while presenting the weaknesses and the strong points of each protocol regards to privacy.

The interest is to arise the real needs of privacy during the execution of the different steps ofk-mean algorithm, thus to discover the best of approaches in case of preserving privacy in k-means algorithm.