Lightweight federated learning for STIs/HIV prediction.

Area of Science:

• Computer Science
• Public Health
• Biomedical Informatics

Background:

• Predicting sexually transmissible infections/human immunodeficiency virus (STIs/HIV) risk is crucial for public health interventions.
• Existing methods often face challenges with data privacy and communication efficiency.
• Federated Learning (FL) offers a decentralized approach to machine learning, preserving data privacy.

Purpose of the Study:

• To develop and evaluate a privacy-preserving model for predicting STIs/HIV risk using Federated Learning.
• To improve communication throughput in distributed machine learning environments for health applications.
• To assess the performance of a Random Forest Federated Learning approach for STIs/HIV risk estimation.

Main Methods:

• Utilized Federated Learning (FL) to train a predictive model across multiple clinics without sharing raw patient data.
• Implemented a Random Forest algorithm within the FL framework for STIs/HIV risk assessment.
• Developed a flexible aggregation process to accommodate varying communication capacities in the system.

Main Results:

• The proposed Random Forest FL model demonstrated significant potential in estimating STIs/HIV risk.
• Achieved superior performance compared to recent studies, with an AUC of 0.97 and high accuracy.
• The flexible aggregation process enhanced system communication throughput.

Conclusions:

• Federated Learning provides a robust solution for privacy-preserving STIs/HIV risk prediction.
• The Random Forest FL approach offers a promising and effective method for improving public health outcomes.
• Future research should explore high-risk populations to further validate the framework's impact.