Collaborative Privacy-Preserving Training of Decision Trees

Abstract

As data generation becomes more ubiquitous, datasets are increasingly spread across several entities. Exploiting such distributed datasets simultaneously, by training machine learning models on them, can be beneficial to many fields, but also raises serious privacy concerns. In this work, a scalable protocol is designed, implemented, and evaluated, for training decision tree models collaboratively across mutually distrustful data providers, while ensuring privacy of each of the stakeholders' data.