EnDuSecFed: an ensemble approach for privacy preserving Federated Learning with dual-security framework for sustainable healthcare
Year: 2026
Type: article
Abstract: 30.0 Recent Artificial Intelligence (AI) improvements have made Machine Learning (ML) very popular for making decisions in healthcare. Machine learning (ML) usually involves processing and collecting data in a single central location, which can lead to security issues. Federated learning (FL) is an advancement that enables training models across multiple clients without collecting all the data in one location. This keeps the original data private because each client uses its data to train a local model and only sends updates (like model weights) to a central server. The server then combines these updates to improve the overall model. However, the decentralized setup of FL can also create security risks that might harm the model’s performance and lead to wrong decisions. This paper suggests a new federated learning system that protects privacy by using Secure Multi-Party Computation (SMPC) for the safe sharing of model updates and an Intrusion Detection System (IDS) at the central server to monitor changes in data. SMPC encrypts client updates to stop unauthorized access and keep local data confidential. Meanwhile, the global IDS identifies unusual updates to keep the overall model strong against attacks. The proposed approach works efficiently on a publicly available healthcare dataset, demonstrating better model security and stability than traditional federated learning systems. The security checks confirm the effectiveness of our approach. Additionally, various ML models like Logistic Regression, Random Forest, Support Vector Classifier, and an ensemble method are implemented and tested in the FL environment for local model training. The results show that the ensemble model performs the best, achieving an accuracy of 99%.
Source: Frontiers in Big Data
Authors: Bela Shrimali, Jenil Gajjar, Swapnoneel Roy, Sanjay Patel, Kanu Patel, Ramesh Ram Naik
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,418
OpenAlex ID: W7125492168