Collaborative Privacy-Preserving Decision Tree Learning

Abstract

Building robust predictive machine learning (ML) models requires access to large datasets. Such datasets can be built by aggregating data held by multiple data providers. An example would be medical datasets, whereby different hospitals hold the health records of different patients. In many scenarios, sharing data among the different data providers is restricted due to privacy issues and/or legal obligations. As a result, the different datasets are exploited individually for training ML models, thus limiting their utility.