Exploring Secure Tree-Based Machine Learning Using Fully Homomorphic Encryption

Abstract

ML algorithms play a vital role in health data analytics in such a way that medical institutions and health practitioners can use them in exploring big data and identifying health trends essential to public health knowledge which would be impossible to do manually. That being said, tree-based algorithms are widely adopted in the healthcare sector especially now that the sector is gradually leaning towards digital transformation. Since they are white box models, they are easy to understand for medical professionals who are not knowledgeable of ML. However, the issue of security and privacy for medical data in health analytics remains prevalent. Data privacy must be preserved in settings when ML is executed in an outsourced ML service provider which may have access in case of unintended data leakage. Hence, our study aims to provide a method for implementing privacy-preserving tree-based ML algorithms by incorporating FHE on medical data, particularly using Concrete ML in a client-server system.