Privacy Preserving Attribute Reduction for Vertically Partitioned Data

Traditional attribute reduction algorithms based on rough set theory assume free access to data. Increasingly, privacy and security constraints may prevent the parties from directly sharing the data and some types of information about the data, thus derailing attribute reduction projects. Distributed attribute reduction, if done correctly, can alleviate this problem. The key is to obtain globally valid attribute reduction result, while providing guarantees on the (non) disclosure of data. In this paper, we consider the problem of computing the attribute reduction of private datasets of two parties, and present a privacy preserving attribute reduction algorithm for vertically partitioned data. The algorithm incorporates secure two-party computation protocol using commutative encryption to minimize the information shared for both semi-honest and malicious environments, while adding little overhead to the relative reduct task.