Differentially private distributed logistic regression using private and public data.

Area of Science:

• Medical Informatics
• Data Privacy
• Machine Learning

Background:

• Privacy protection is critical in medical informatics.
• Differential privacy offers provable privacy but can reduce data utility due to noise.
• Publicly available medical data can be leveraged to improve privacy-preserving methods.

Purpose of the Study:

• To develop a differentially private distributed logistic regression model that effectively utilizes both public and private datasets.
• To decrease the amount of noise introduced by differentially private methods by incorporating public data.
• To improve the utility of privacy-preserving machine learning models in medical research.

Main Methods:

• Modification of the Newton-Raphson method's update step.
• Development of a distributed logistic regression model incorporating both public and private data.
• Implementation of differential privacy techniques within the model.

Main Results:

• The proposed algorithm was tested on three datasets.
• The new model demonstrated superior performance compared to models using only public or only private data.
• The algorithm showed advantages across various scenarios.

Conclusions:

• Logistic regression models built with the new algorithm offer improved utility.
• The approach successfully integrates private and public datasets without compromising privacy guarantees.
• This method provides a better balance between privacy and utility in medical data analysis.