A new method for preserving data privacy based on the non-negative matrix factorization clustering

Companies are increasingly concerned about the disclosure and violation of users' privacy, and this has led many researchers to focus on developing of data privacy methods. These methods affect the original data and publish the data in a different form by keeping their features and relationships. This paper suggests an algorithm to generate data for publishing with different privacy level for a given original data set based on non-negative matrix factorization clustering. Implementation results with two different approaches to various standard data sets show that our proposed algorithm can satisfy the original data constraints in addition to generating data with different privacy levels. The method proposed in this article is implemented on high-dimensional datasets, while some fuzzy-based microaggregation methods cannot be implemented on them. Also, the experimental results using fuzzy $c$-mean show that the information loss is very small. Therefore, the proposed algorithm can publish data that can be relied on by legal users while preserving privacy.