Privacy-Preserving Collaborative Prediction using Random Forests.

Area of Science:

• Computer Science
• Machine Learning
• Data Privacy

Background:

• Privacy-preserving machine learning (PPML) is crucial for collaborative data analysis, especially in sensitive domains like healthcare.
• Existing methods struggle to balance data sharing needs with stringent patient privacy requirements.

Purpose of the Study:

• To develop a novel PPML approach for ensemble methods, specifically random forests.
• To enable secure collaborative model training using distributed, private datasets.

Main Methods:

• Proposed a decentralized approach where each entity trains a local model on its private data.
• Developed a secure aggregation method for predictions from local models without revealing individual data.

Main Results:

• Implemented the PPML approach for random forests.
• Demonstrated high efficiency and potential accuracy benefits through experiments on real-world datasets, including electronic health records (EHR).

Conclusions:

• The proposed PPML method effectively addresses privacy concerns in ensemble learning.
• This approach facilitates secure collaborative analysis of sensitive data, enhancing predictive modeling capabilities.