Privacy preserving RBF kernel support vector machine.

Area of Science:

• Health Informatics
• Biomedical Data Science
• Machine Learning

Background:

• Data sharing is crucial for healthcare research but faces privacy challenges.
• Existing privacy-preserving methods, like differential privacy, often introduce excessive noise in biomedical data analysis.
• Current approaches do not fully utilize available data characteristics, impacting utility.

Purpose of the Study:

• To develop a hybrid privacy-preserving model that integrates public and private data for improved healthcare research utility.
• To enhance differentially private machine learning models by incorporating open-consented data without sacrificing privacy guarantees.
• To address the limitations of existing methods in balancing privacy and data utility in biomedical research.

Main Methods:

• Developed a hybrid differentially private Support Vector Machine (SVM) model.
• Incorporated a Radial Basis Function (RBF) kernel to handle nonlinearly separable data.
• Utilized both open-consented public data and sensitive private data within the model framework.

Main Results:

• The hybrid model demonstrated superior performance compared to SVMs using only public data or differentially private SVMs trained solely on private data.
• The proposed method achieved performance metrics comparable to non-private SVMs trained on private data.
• The approach effectively improved data utility while maintaining rigorous privacy standards.

Conclusions:

• A hybrid approach combining public and private data significantly enhances the utility of privacy-preserving machine learning in healthcare research.
• This method offers a practical solution for leveraging diverse data sources while adhering to differential privacy standards.
• The developed differentially private SVM model provides a robust framework for secure and effective biomedical data analysis.