A Robust Pipeline for Differentially Private Federated Learning on Imbalanced Clinical Data using SMOTETomek and FedProx

2025-08-06 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Data Science & Analytics · Depth: Advanced, quick

Summary

A robust pipeline for differentially private federated learning (FL) has been developed to address the challenges of imbalanced clinical data in collaborative health research. This framework, detailed in a study accepted by the Journal of the Brazilian Computer Society, focuses on cardiovascular risk prediction. Initial experiments with standard FL methods on imbalanced data resulted in a recall of zero. To overcome this, the pipeline integrates the hybrid Synthetic Minority Over-sampling Technique with Tomek Links (SMOTETomek) at the client level, successfully creating a clinically useful model. Further optimization for non-IID data was achieved using a tuned FedProx algorithm, which consistently outperformed standard FedAvg. The research identified an optimal operational region on the privacy-utility frontier, demonstrating that strong privacy guarantees (with epsilon 9.0) can be maintained alongside high clinical utility (recall greater than 77%). The study includes 3 figures and 7 tables.

Key takeaway

For AI Scientists and Research Scientists developing secure diagnostic tools with federated learning on clinical data, you should integrate client-side SMOTETomek to address class imbalance and employ a tuned FedProx algorithm for non-IID data. This approach allows you to achieve strong differential privacy (epsilon 9.0) while maintaining high clinical utility (recall > 77%), overcoming the privacy-utility trade-off often seen in medical AI. Consider this blueprint for robust, secure, and accurate real-world healthcare applications.

Key insights

Combining SMOTETomek and FedProx enables differentially private federated learning on imbalanced clinical data with high utility.

Principles

• Class imbalance severely degrades FL model recall.
• Privacy budget (epsilon) has a non-linear trade-off with utility.
• Optimized FedProx outperforms FedAvg for non-IID data.

Method

Implement client-side SMOTETomek for data balancing, then optimize FL with tuned FedProx for non-IID data, and identify optimal privacy-utility frontier.

In practice

• Apply SMOTETomek to client data before FL training.
• Tune FedProx for non-IID medical datasets.
• Target epsilon 9.0 for strong privacy with high recall.

Topics

• Federated Learning
• Differential Privacy
• SMOTETomek
• FedProx
• Clinical Data
• Cardiovascular Risk Prediction
• Data Imbalance

Best for: AI Scientist, Research Scientist, AI Security Engineer

Related on AIssential

• Statistical Limits and Efficient Algorithms for Differentially Private Federated Learning — New methods improve federated learning accuracy and communication efficiency while preserving differential privacy.
• Federated learning with swarm intelligence for efficient and secure medical image analysis — Swarm intelligence-enhanced federated learning improves medical image analysis accuracy and efficiency while preserving privacy.
• Class-Grouped Normalized Momentum and Faster Hyperparameter Exploration to Tackle Class Imbalance in Federated Learning — FedCGNM tackles federated learning class imbalance via client-side class-grouped normalized momentum, enhanced by FedHOO for hyperparameter optimization.
• Federated Survival Analysis in Healthcare: A Multi-Model Evaluation on Cross-Institutional Heterogeneous Breast Cancer Data — Federated learning enables robust survival modeling on private, heterogeneous healthcare data, with Random Survival Forest (RSF) offering the best balance of performance.
• FedXDS: Leveraging Model Attribution Methods to counteract Data Heterogeneity in Federated Learning — FedXDS uses XAI feature attribution to selectively share data, mitigating heterogeneity in federated learning while preserving privacy.
• Federated Deep Learning for Privacy-Preserving Cardiovascular Disease Risk Prediction — Federated deep learning improves cardiovascular risk prediction across heterogeneous datasets while preserving patient privacy.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.