Privacy preserving distributed extremely randomized trees

Abstract

Applying machine learning and data mining algorithms over data distributed in multiple sources is challenging. One complication is to perform data analysis without compromising personal information, which is a primary concern in healthcare applications. Another issue involves communication overhead incurred from the transfer of raw data from one party to others for conducting centralized data mining. In healthcare applications, we are particularly interested in running data mining algorithms over big data without disclosing sensitive information about data subjects due to privacy and legal concerns. In this paper, we consider the classification problem and show how the Extremely Randomized Trees (ERT) algorithm could be adapted for settings where (structured) data is distributed over multiple sources. We propose the Privacy-Preserving Distributed ERT approach for privacy-preserving utilization of the ERT algorithm in a distributed setting. To the best of our knowledge, this is the first application of the ERT algorithm in the distributed setting, with privacy consideration (without sharing the raw data or intermediate training values), without any loss in classification performance.